ML_DetectionFraud.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "UNIVERSIDADE ESTADUAL DO CEARÁ \\\n",
    "MESTRADO ACADÊMICO EM CIÊNCIA DA COMPUTAÇÃO \\\n",
    "MINERAÇÃO MASSIVA DE DADOS\n",
    "\n",
    "Daniel Gleison Moreira Lira \\\n",
    "daniel.gleison@aluno.uece.br\n",
    "\n",
    "---\n",
    "# Mecanismo de predição de fraudes financeiras utilizando aprendizado de máquina e processamento distribuído\n",
    "---\n",
    "\n",
    "\n",
    "## Dataset\n",
    "\n",
    "https://www.kaggle.com/ntnu-testimon/paysim1\n",
    "\n",
    "PS_20174392719_1491204439457_log.csv\\\n",
    "Date created: 2017-03-31\n",
    "\n",
    "### Predicted attribute:\n",
    "Class of fraud detection\n",
    "\n",
    "### Number of Instances:\n",
    "6.353.307\n",
    "\n",
    "### Number of Attributes:\n",
    "11 attributes (5 Decimal, 3 Integer, 3 String)\n",
    "\n",
    "### Attribute Information:\n",
    "\n",
    "This is a sample of 1 row with headers explanation:\n",
    "\n",
    "1,PAYMENT,1060.31,C429214117,1089.0,28.69,M1591654462,0.0,0.0,0,0\n",
    "\n",
    "step - maps a unit of time in the real world. In this case 1 step is 1 hour of time. Total steps 744 (30 days simulation).\n",
    "\n",
    "type - CASH-IN, CASH-OUT, DEBIT, PAYMENT and TRANSFER.\n",
    "\n",
    "amount - amount of the transaction in local currency.\n",
    "\n",
    "nameOrig - customer who started the transaction\n",
    "\n",
    "oldbalanceOrg - initial balance before the transaction\n",
    "\n",
    "newbalanceOrig - new balance after the transaction\n",
    "\n",
    "nameDest - customer who is the recipient of the transaction\n",
    "\n",
    "oldbalanceDest - initial balance recipient before the transaction. Note that there is not information for customers that start with M (Merchants).\n",
    "\n",
    "newbalanceDest - new balance recipient after the transaction. Note that there is not information for customers that start with M (Merchants).\n",
    "\n",
    "isFraud - This is the transactions made by the fraudulent agents inside the simulation. In this specific dataset the fraudulent behavior of the agents aims to profit by taking control or customers accounts and try to empty the funds by transferring to another account and then cashing out of the system.\n",
    "\n",
    "isFlaggedFraud - The business model aims to control massive transfers from one account to another and flags illegal attempts. An illegal attempt in this dataset is an attempt to transfer more than 200.000 in a single transaction.\n",
    "\n",
    "### Missing Attribute Values: \n",
    "None\n",
    "\n",
    "### Class Distribution: \n",
    "2 Classes \\\n",
    "6.354.407 without fraud and 8.213 with fraud\n",
    "\n",
    "---"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Spark Lib\n",
    "import findspark\n",
    "findspark.init()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load libraries\n",
    "import pyspark\n",
    "from pyspark.sql import SparkSession\n",
    "from pyspark.sql.functions import isnull, when, count, col\n",
    "\n",
    "from pyspark.ml import Pipeline\n",
    "from pyspark.ml.classification import DecisionTreeClassifier\n",
    "from pyspark.ml.feature import StringIndexer, VectorIndexer\n",
    "from pyspark.ml.evaluation import MulticlassClassificationEvaluator\n",
    "from pyspark.mllib.util import MLUtils\n",
    "\n",
    "from pyspark.ml.feature import StringIndexer, IndexToString\n",
    "from pyspark.ml.feature import VectorAssembler, VectorIndexer\n",
    "\n",
    "from pyspark.ml.classification import DecisionTreeClassifier\n",
    "from pyspark.ml.classification import RandomForestClassifier\n",
    "from pyspark.ml.classification import MultilayerPerceptronClassifier\n",
    "from pyspark.ml.classification import NaiveBayes\n",
    "from pyspark.ml.classification import LogisticRegression\n",
    "from pyspark.ml.classification import LinearSVC, OneVsRest\n",
    "\n",
    "from pyspark.ml.evaluation import MulticlassClassificationEvaluator, BinaryClassificationEvaluator\n",
    "\n",
    "from pyspark.ml.linalg import Vectors\n",
    "from pyspark.mllib.util import MLUtils\n",
    "\n",
    "## SKLearn Lib\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "from sklearn.preprocessing import LabelEncoder\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.neighbors import KNeighborsClassifier\n",
    "from sklearn.metrics import confusion_matrix, plot_confusion_matrix, accuracy_score\n",
    "from sklearn.model_selection import cross_val_score\n",
    "from joblib import dump, load\n",
    "import pickle\n",
    "\n",
    "\n",
    "import time\n",
    "start_time = time.time()\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Criação do ambiente Spark"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "            <div>\n",
       "                <p><b>SparkSession - in-memory</b></p>\n",
       "                \n",
       "        <div>\n",
       "            <p><b>SparkContext</b></p>\n",
       "\n",
       "            <p><a href=\"http://10.129.64.20:4040\">Spark UI</a></p>\n",
       "\n",
       "            <dl>\n",
       "              <dt>Version</dt>\n",
       "                <dd><code>v2.4.5</code></dd>\n",
       "              <dt>Master</dt>\n",
       "                <dd><code>local[*]</code></dd>\n",
       "              <dt>AppName</dt>\n",
       "                <dd><code>MachineLearningFraud</code></dd>\n",
       "            </dl>\n",
       "        </div>\n",
       "        \n",
       "            </div>\n",
       "        "
      ],
      "text/plain": [
       "<pyspark.sql.session.SparkSession at 0x7f7ddbcc5b00>"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "spark = SparkSession.builder \\\n",
    "        .master(\"local[*]\") \\\n",
    "        .appName(\"MachineLearningFraud\") \\\n",
    "        .getOrCreate()\n",
    "spark"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Importação e compreensão dos dados"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+----+--------+---------+-----------+-------------+--------------+-----------+--------------+--------------+-------+--------------+\n",
      "|step|    type|   amount|   nameOrig|oldbalanceOrg|newbalanceOrig|   nameDest|oldbalanceDest|newbalanceDest|isFraud|isFlaggedFraud|\n",
      "+----+--------+---------+-----------+-------------+--------------+-----------+--------------+--------------+-------+--------------+\n",
      "|   1| PAYMENT|  9839.64|C1231006815|     170136.0|     160296.36|M1979787155|           0.0|           0.0|      0|             0|\n",
      "|   1| PAYMENT|  1864.28|C1666544295|      21249.0|      19384.72|M2044282225|           0.0|           0.0|      0|             0|\n",
      "|   1|TRANSFER|    181.0|C1305486145|        181.0|           0.0| C553264065|           0.0|           0.0|      1|             0|\n",
      "|   1|CASH_OUT|    181.0| C840083671|        181.0|           0.0|  C38997010|       21182.0|           0.0|      1|             0|\n",
      "|   1| PAYMENT| 11668.14|C2048537720|      41554.0|      29885.86|M1230701703|           0.0|           0.0|      0|             0|\n",
      "|   1| PAYMENT|  7817.71|  C90045638|      53860.0|      46042.29| M573487274|           0.0|           0.0|      0|             0|\n",
      "|   1| PAYMENT|  7107.77| C154988899|     183195.0|     176087.23| M408069119|           0.0|           0.0|      0|             0|\n",
      "|   1| PAYMENT|  7861.64|C1912850431|    176087.23|     168225.59| M633326333|           0.0|           0.0|      0|             0|\n",
      "|   1| PAYMENT|  4024.36|C1265012928|       2671.0|           0.0|M1176932104|           0.0|           0.0|      0|             0|\n",
      "|   1|   DEBIT|  5337.77| C712410124|      41720.0|      36382.23| C195600860|       41898.0|      40348.79|      0|             0|\n",
      "|   1|   DEBIT|  9644.94|C1900366749|       4465.0|           0.0| C997608398|       10845.0|     157982.12|      0|             0|\n",
      "|   1| PAYMENT|  3099.97| C249177573|      20771.0|      17671.03|M2096539129|           0.0|           0.0|      0|             0|\n",
      "|   1| PAYMENT|  2560.74|C1648232591|       5070.0|       2509.26| M972865270|           0.0|           0.0|      0|             0|\n",
      "|   1| PAYMENT| 11633.76|C1716932897|      10127.0|           0.0| M801569151|           0.0|           0.0|      0|             0|\n",
      "|   1| PAYMENT|  4098.78|C1026483832|     503264.0|     499165.22|M1635378213|           0.0|           0.0|      0|             0|\n",
      "|   1|CASH_OUT|229133.94| C905080434|      15325.0|           0.0| C476402209|        5083.0|      51513.44|      0|             0|\n",
      "|   1| PAYMENT|  1563.82| C761750706|        450.0|           0.0|M1731217984|           0.0|           0.0|      0|             0|\n",
      "|   1| PAYMENT|  1157.86|C1237762639|      21156.0|      19998.14|M1877062907|           0.0|           0.0|      0|             0|\n",
      "|   1| PAYMENT|   671.64|C2033524545|      15123.0|      14451.36| M473053293|           0.0|           0.0|      0|             0|\n",
      "|   1|TRANSFER| 215310.3|C1670993182|        705.0|           0.0|C1100439041|       22425.0|           0.0|      0|             0|\n",
      "+----+--------+---------+-----------+-------------+--------------+-----------+--------------+--------------+-------+--------------+\n",
      "only showing top 20 rows\n",
      "\n"
     ]
    }
   ],
   "source": [
    "data_path='./data/'\n",
    "df_original = spark.read.format('csv')\\\n",
    "                   .options(sep=',',header='true',inferschema='true').\\\n",
    "                   load(data_path+'PS_20174392719_1491204439457_log.csv')\n",
    "df_original.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Balanceamento do dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+-------+-------+\n",
      "|isFraud|  count|\n",
      "+-------+-------+\n",
      "|      0|6354407|\n",
      "+-------+-------+\n",
      "\n",
      "+-------+-----+\n",
      "|isFraud|count|\n",
      "+-------+-----+\n",
      "|      1| 8213|\n",
      "+-------+-----+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "df0 = df_original[df_original.isFraud==0]\n",
    "df1 = df_original[df_original.isFraud==1]\n",
    "df0.groupby('isFraud').count().show()\n",
    "df1.groupby('isFraud').count().show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+-------+-----+\n",
      "|isFraud|count|\n",
      "+-------+-----+\n",
      "|      0| 8213|\n",
      "+-------+-----+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "guessedFraction = 1.0\n",
    "noOfSamples = df1.count()\n",
    "df0 = df0.sample(True, guessedFraction).limit(noOfSamples)\n",
    "df0.groupby('isFraud').count().show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "16426"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_balanceado = df0.union(df1)\n",
    "df_balanceado.count()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Análise exploratória dos dados"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "df_original = df_balanceado"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "root\n",
      " |-- step: integer (nullable = true)\n",
      " |-- type: string (nullable = true)\n",
      " |-- amount: double (nullable = true)\n",
      " |-- nameOrig: string (nullable = true)\n",
      " |-- oldbalanceOrg: double (nullable = true)\n",
      " |-- newbalanceOrig: double (nullable = true)\n",
      " |-- nameDest: string (nullable = true)\n",
      " |-- oldbalanceDest: double (nullable = true)\n",
      " |-- newbalanceDest: double (nullable = true)\n",
      " |-- isFraud: integer (nullable = true)\n",
      " |-- isFlaggedFraud: integer (nullable = true)\n",
      "\n"
     ]
    }
   ],
   "source": [
    "df_original.printSchema()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "16426"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_original.count()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+-------+-----+\n",
      "|IsFraud|count|\n",
      "+-------+-----+\n",
      "|      1| 8213|\n",
      "|      0| 8213|\n",
      "+-------+-----+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "df_original.groupBy('IsFraud').count().show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+--------------+-----+\n",
      "|isFlaggedFraud|count|\n",
      "+--------------+-----+\n",
      "|             1|   16|\n",
      "|             0|16410|\n",
      "+--------------+-----+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "df_original.groupBy('isFlaggedFraud').count().show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Estatística descritiva"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>0</th>\n",
       "      <th>1</th>\n",
       "      <th>2</th>\n",
       "      <th>3</th>\n",
       "      <th>4</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>summary</th>\n",
       "      <td>count</td>\n",
       "      <td>mean</td>\n",
       "      <td>stddev</td>\n",
       "      <td>min</td>\n",
       "      <td>max</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>step</th>\n",
       "      <td>16426</td>\n",
       "      <td>186.0233166930476</td>\n",
       "      <td>238.0792153027866</td>\n",
       "      <td>1</td>\n",
       "      <td>743</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>type</th>\n",
       "      <td>16426</td>\n",
       "      <td>None</td>\n",
       "      <td>None</td>\n",
       "      <td>CASH_IN</td>\n",
       "      <td>TRANSFER</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>amount</th>\n",
       "      <td>16426</td>\n",
       "      <td>784727.7085468163</td>\n",
       "      <td>1839630.8982522383</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0E7</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>nameOrig</th>\n",
       "      <td>16426</td>\n",
       "      <td>None</td>\n",
       "      <td>None</td>\n",
       "      <td>C1000036340</td>\n",
       "      <td>C999864329</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>oldbalanceOrg</th>\n",
       "      <td>16426</td>\n",
       "      <td>1284608.1605466937</td>\n",
       "      <td>2969170.076606432</td>\n",
       "      <td>0.0</td>\n",
       "      <td>5.958504037E7</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>newbalanceOrig</th>\n",
       "      <td>16426</td>\n",
       "      <td>569293.4797577027</td>\n",
       "      <td>2140225.0867706854</td>\n",
       "      <td>0.0</td>\n",
       "      <td>4.958504037E7</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>nameDest</th>\n",
       "      <td>16426</td>\n",
       "      <td>None</td>\n",
       "      <td>None</td>\n",
       "      <td>C1000039615</td>\n",
       "      <td>M999221400</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>oldbalanceDest</th>\n",
       "      <td>16426</td>\n",
       "      <td>778124.3870461485</td>\n",
       "      <td>3081242.623673352</td>\n",
       "      <td>0.0</td>\n",
       "      <td>2.3623051682E8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>newbalanceDest</th>\n",
       "      <td>16426</td>\n",
       "      <td>1237570.697896019</td>\n",
       "      <td>3566922.0427193088</td>\n",
       "      <td>0.0</td>\n",
       "      <td>2.3672649466E8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>isFraud</th>\n",
       "      <td>16426</td>\n",
       "      <td>0.5</td>\n",
       "      <td>0.5000152204684895</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>isFlaggedFraud</th>\n",
       "      <td>16426</td>\n",
       "      <td>9.740655059052721E-4</td>\n",
       "      <td>0.031195768116294764</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                    0                     1                     2  \\\n",
       "summary         count                  mean                stddev   \n",
       "step            16426     186.0233166930476     238.0792153027866   \n",
       "type            16426                  None                  None   \n",
       "amount          16426     784727.7085468163    1839630.8982522383   \n",
       "nameOrig        16426                  None                  None   \n",
       "oldbalanceOrg   16426    1284608.1605466937     2969170.076606432   \n",
       "newbalanceOrig  16426     569293.4797577027    2140225.0867706854   \n",
       "nameDest        16426                  None                  None   \n",
       "oldbalanceDest  16426     778124.3870461485     3081242.623673352   \n",
       "newbalanceDest  16426     1237570.697896019    3566922.0427193088   \n",
       "isFraud         16426                   0.5    0.5000152204684895   \n",
       "isFlaggedFraud  16426  9.740655059052721E-4  0.031195768116294764   \n",
       "\n",
       "                          3               4  \n",
       "summary                 min             max  \n",
       "step                      1             743  \n",
       "type                CASH_IN        TRANSFER  \n",
       "amount                  0.0           1.0E7  \n",
       "nameOrig        C1000036340      C999864329  \n",
       "oldbalanceOrg           0.0   5.958504037E7  \n",
       "newbalanceOrig          0.0   4.958504037E7  \n",
       "nameDest        C1000039615      M999221400  \n",
       "oldbalanceDest          0.0  2.3623051682E8  \n",
       "newbalanceDest          0.0  2.3672649466E8  \n",
       "isFraud                   0               1  \n",
       "isFlaggedFraud            0               1  "
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_original.describe().toPandas().transpose()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Identificação de valores ausentes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+----+----+------+--------+-------------+--------------+--------+--------------+--------------+-------+--------------+\n",
      "|step|type|amount|nameOrig|oldbalanceOrg|newbalanceOrig|nameDest|oldbalanceDest|newbalanceDest|isFraud|isFlaggedFraud|\n",
      "+----+----+------+--------+-------------+--------------+--------+--------------+--------------+-------+--------------+\n",
      "|   0|   0|     0|       0|            0|             0|       0|             0|             0|      0|             0|\n",
      "+----+----+------+--------+-------------+--------------+--------+--------------+--------------+-------+--------------+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "df_original.select([count(when(isnull(c), c)).alias(c) for c in df_original.columns]).show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Matrix de correlação"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>step</th>\n",
       "      <th>amount</th>\n",
       "      <th>oldbalanceOrg</th>\n",
       "      <th>newbalanceOrig</th>\n",
       "      <th>oldbalanceDest</th>\n",
       "      <th>newbalanceDest</th>\n",
       "      <th>isFraud</th>\n",
       "      <th>isFlaggedFraud</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>step</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.313246</td>\n",
       "      <td>0.128546</td>\n",
       "      <td>-0.110535</td>\n",
       "      <td>-0.048694</td>\n",
       "      <td>0.023170</td>\n",
       "      <td>0.766115</td>\n",
       "      <td>0.046107</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>amount</th>\n",
       "      <td>0.313246</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.721365</td>\n",
       "      <td>0.145564</td>\n",
       "      <td>-0.007684</td>\n",
       "      <td>0.247215</td>\n",
       "      <td>0.371412</td>\n",
       "      <td>0.069201</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>oldbalanceOrg</th>\n",
       "      <td>0.128546</td>\n",
       "      <td>0.721365</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.784466</td>\n",
       "      <td>0.057673</td>\n",
       "      <td>0.187867</td>\n",
       "      <td>0.122954</td>\n",
       "      <td>0.068709</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>newbalanceOrig</th>\n",
       "      <td>-0.110535</td>\n",
       "      <td>0.145564</td>\n",
       "      <td>0.784466</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.115900</td>\n",
       "      <td>0.075829</td>\n",
       "      <td>-0.176109</td>\n",
       "      <td>0.105758</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>oldbalanceDest</th>\n",
       "      <td>-0.048694</td>\n",
       "      <td>-0.007684</td>\n",
       "      <td>0.057673</td>\n",
       "      <td>0.115900</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.898623</td>\n",
       "      <td>-0.075905</td>\n",
       "      <td>-0.007886</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>newbalanceDest</th>\n",
       "      <td>0.023170</td>\n",
       "      <td>0.247215</td>\n",
       "      <td>0.187867</td>\n",
       "      <td>0.075829</td>\n",
       "      <td>0.898623</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.011814</td>\n",
       "      <td>-0.010834</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>isFraud</th>\n",
       "      <td>0.766115</td>\n",
       "      <td>0.371412</td>\n",
       "      <td>0.122954</td>\n",
       "      <td>-0.176109</td>\n",
       "      <td>-0.075905</td>\n",
       "      <td>0.011814</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.031225</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>isFlaggedFraud</th>\n",
       "      <td>0.046107</td>\n",
       "      <td>0.069201</td>\n",
       "      <td>0.068709</td>\n",
       "      <td>0.105758</td>\n",
       "      <td>-0.007886</td>\n",
       "      <td>-0.010834</td>\n",
       "      <td>0.031225</td>\n",
       "      <td>1.000000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                    step    amount  oldbalanceOrg  newbalanceOrig  \\\n",
       "step            1.000000  0.313246       0.128546       -0.110535   \n",
       "amount          0.313246  1.000000       0.721365        0.145564   \n",
       "oldbalanceOrg   0.128546  0.721365       1.000000        0.784466   \n",
       "newbalanceOrig -0.110535  0.145564       0.784466        1.000000   \n",
       "oldbalanceDest -0.048694 -0.007684       0.057673        0.115900   \n",
       "newbalanceDest  0.023170  0.247215       0.187867        0.075829   \n",
       "isFraud         0.766115  0.371412       0.122954       -0.176109   \n",
       "isFlaggedFraud  0.046107  0.069201       0.068709        0.105758   \n",
       "\n",
       "                oldbalanceDest  newbalanceDest   isFraud  isFlaggedFraud  \n",
       "step                 -0.048694        0.023170  0.766115        0.046107  \n",
       "amount               -0.007684        0.247215  0.371412        0.069201  \n",
       "oldbalanceOrg         0.057673        0.187867  0.122954        0.068709  \n",
       "newbalanceOrig        0.115900        0.075829 -0.176109        0.105758  \n",
       "oldbalanceDest        1.000000        0.898623 -0.075905       -0.007886  \n",
       "newbalanceDest        0.898623        1.000000  0.011814       -0.010834  \n",
       "isFraud              -0.075905        0.011814  1.000000        0.031225  \n",
       "isFlaggedFraud       -0.007886       -0.010834  0.031225        1.000000  "
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_original.toPandas().corr()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Transformação do dataset"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Indexação dos atributos"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+----+-------+--------+-----------+-------------+--------------+-----------+--------------+--------------+-------+--------------+---------+-----+\n",
      "|step|   type|  amount|   nameOrig|oldbalanceOrg|newbalanceOrig|   nameDest|oldbalanceDest|newbalanceDest|isFraud|isFlaggedFraud|indexType|label|\n",
      "+----+-------+--------+-----------+-------------+--------------+-----------+--------------+--------------+-------+--------------+---------+-----+\n",
      "|   1|PAYMENT| 1864.28|C1666544295|      21249.0|      19384.72|M2044282225|           0.0|           0.0|      0|             0|      2.0|  0.0|\n",
      "|   1|PAYMENT|11668.14|C2048537720|      41554.0|      29885.86|M1230701703|           0.0|           0.0|      0|             0|      2.0|  0.0|\n",
      "|   1|PAYMENT|11668.14|C2048537720|      41554.0|      29885.86|M1230701703|           0.0|           0.0|      0|             0|      2.0|  0.0|\n",
      "|   1|PAYMENT| 7107.77| C154988899|     183195.0|     176087.23| M408069119|           0.0|           0.0|      0|             0|      2.0|  0.0|\n",
      "|   1|PAYMENT| 7107.77| C154988899|     183195.0|     176087.23| M408069119|           0.0|           0.0|      0|             0|      2.0|  0.0|\n",
      "+----+-------+--------+-----------+-------------+--------------+-----------+--------------+--------------+-------+--------------+---------+-----+\n",
      "only showing top 5 rows\n",
      "\n"
     ]
    }
   ],
   "source": [
    "indexer = StringIndexer(inputCol='type', outputCol='indexType').fit(df_original)\n",
    "df_indexado = indexer.transform(df_original)\n",
    "\n",
    "indexer = StringIndexer(inputCol='isFraud', outputCol='label').fit(df_original)\n",
    "df_indexado = indexer.transform(df_indexado)\n",
    "\n",
    "labelReverse = IndexToString().setInputCol('label')\n",
    "indexTypeReverse = IndexToString().setInputCol('indexType')\n",
    "\n",
    "df_indexado.show(5)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Exclusão de atributos"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+----+--------+-------------+--------------+--------------+--------------+---------+-----+\n",
      "|step|  amount|oldbalanceOrg|newbalanceOrig|oldbalanceDest|newbalanceDest|indexType|label|\n",
      "+----+--------+-------------+--------------+--------------+--------------+---------+-----+\n",
      "|   1| 1864.28|      21249.0|      19384.72|           0.0|           0.0|      2.0|  0.0|\n",
      "|   1|11668.14|      41554.0|      29885.86|           0.0|           0.0|      2.0|  0.0|\n",
      "|   1|11668.14|      41554.0|      29885.86|           0.0|           0.0|      2.0|  0.0|\n",
      "|   1| 7107.77|     183195.0|     176087.23|           0.0|           0.0|      2.0|  0.0|\n",
      "|   1| 7107.77|     183195.0|     176087.23|           0.0|           0.0|      2.0|  0.0|\n",
      "+----+--------+-------------+--------------+--------------+--------------+---------+-----+\n",
      "only showing top 5 rows\n",
      "\n"
     ]
    }
   ],
   "source": [
    "df_selecionado = df_indexado.drop('type','nameOrig','nameDest','isFraud','isFlaggedFraud')\n",
    "df_selecionado.show(5)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Seleção dos atributos"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+----+--------+-------------+--------------+--------------+--------------+---------+-----+\n",
      "|step|  amount|oldbalanceOrg|newbalanceOrig|oldbalanceDest|newbalanceDest|indexType|label|\n",
      "+----+--------+-------------+--------------+--------------+--------------+---------+-----+\n",
      "|   1| 1864.28|      21249.0|      19384.72|           0.0|           0.0|      2.0|  0.0|\n",
      "|   1|11668.14|      41554.0|      29885.86|           0.0|           0.0|      2.0|  0.0|\n",
      "|   1|11668.14|      41554.0|      29885.86|           0.0|           0.0|      2.0|  0.0|\n",
      "|   1| 7107.77|     183195.0|     176087.23|           0.0|           0.0|      2.0|  0.0|\n",
      "|   1| 7107.77|     183195.0|     176087.23|           0.0|           0.0|      2.0|  0.0|\n",
      "+----+--------+-------------+--------------+--------------+--------------+---------+-----+\n",
      "only showing top 5 rows\n",
      "\n"
     ]
    }
   ],
   "source": [
    "df_selecionado.show(5)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Criação da matrix de classificação"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+-----+--------------------------------------------+\n",
      "|label|features                                    |\n",
      "+-----+--------------------------------------------+\n",
      "|0.0  |[1.0,1864.28,21249.0,19384.72,0.0,0.0,2.0]  |\n",
      "|0.0  |[1.0,11668.14,41554.0,29885.86,0.0,0.0,2.0] |\n",
      "|0.0  |[1.0,11668.14,41554.0,29885.86,0.0,0.0,2.0] |\n",
      "|0.0  |[1.0,7107.77,183195.0,176087.23,0.0,0.0,2.0]|\n",
      "|0.0  |[1.0,7107.77,183195.0,176087.23,0.0,0.0,2.0]|\n",
      "+-----+--------------------------------------------+\n",
      "only showing top 5 rows\n",
      "\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "16426"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ignore = ['label']\n",
    "list = [x for x in df_selecionado.columns if x not in ignore]\n",
    "\n",
    "assembler = VectorAssembler(\n",
    "            inputCols= list,\n",
    "            outputCol='features')\n",
    "\n",
    "df_transformado = (assembler.transform(df_selecionado).select('label','features'))\n",
    "df_transformado.show(truncate = False, n = 5)\n",
    "df_transformado.count()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Divisão do dataset para treinamento e teste"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Percentual da base de treinamento 70.0 %\n",
      "Percentual da base de teste 30.0 %\n",
      "Quantidade de registros da base de treinamento: 11559\n",
      "Quantidade de registros da base de treinamento: 4867\n"
     ]
    }
   ],
   "source": [
    "train_sample = 0.7\n",
    "test_sample = 0.3\n",
    "seed = 1234\n",
    "\n",
    "(train, test) = df_transformado.randomSplit([train_sample, test_sample],seed)\n",
    "\n",
    "num_train = df_transformado.count() * train_sample\n",
    "num_test = df_transformado.count() * test_sample\n",
    "\n",
    "print('Percentual da base de treinamento', train_sample*100, '%')\n",
    "print('Percentual da base de teste', test_sample*100, '%')\n",
    "print('Quantidade de registros da base de treinamento:', train.count())\n",
    "print('Quantidade de registros da base de treinamento:', test.count())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+-----+-----+\n",
      "|label|count|\n",
      "+-----+-----+\n",
      "|  0.0| 5773|\n",
      "|  1.0| 5786|\n",
      "+-----+-----+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "train.groupby('label').count().show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+-----+-----+\n",
      "|label|count|\n",
      "+-----+-----+\n",
      "|  0.0| 2440|\n",
      "|  1.0| 2427|\n",
      "+-----+-----+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "test.groupby('label').count().show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Treinamento, teste e avaliação dos modelos de predição"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Decision Tree (DT)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Treinamento do modelo de predição\n",
    "start_time = time.time()\n",
    "trainer = DecisionTreeClassifier(featuresCol='features', labelCol='label', predictionCol='prediction', probabilityCol='probability',\\\n",
    "                                 rawPredictionCol='rawPrediction', maxDepth=5, maxBins=32, minInstancesPerNode=1, minInfoGain=0.0,\\\n",
    "                                 maxMemoryInMB=256, cacheNodeIds=False, checkpointInterval=10, impurity='gini', seed=None)\n",
    "model_dt = trainer.fit(train)\n",
    "time_dt_train = time.time() - start_time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Execução do modelo de predição na base de teste\n",
    "start_time = time.time()\n",
    "result_dt = model_dt.transform(test)\n",
    "time_dt_pred = time.time() - start_time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Cálculo da acurácia do modelo de predição\n",
    "evaluator = MulticlassClassificationEvaluator(labelCol='label', predictionCol='prediction',\\\n",
    "            metricName='accuracy')\n",
    "accuracy_dt = evaluator.evaluate(result_dt) * 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Cálculo do recall do modelo de predição\n",
    "evaluator = MulticlassClassificationEvaluator(labelCol='label', predictionCol='prediction',\\\n",
    "            metricName='weightedRecall')\n",
    "recall_dt = evaluator.evaluate(result_dt) * 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Cálculo da precisão do modelo de predição\n",
    "evaluator = MulticlassClassificationEvaluator(labelCol='label', predictionCol='prediction',\\\n",
    "            metricName='weightedPrecision')\n",
    "precision_dt = evaluator.evaluate(result_dt) * 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Cálculo da F1 score do modelo de predição\n",
    "evaluator = MulticlassClassificationEvaluator(labelCol='label', predictionCol='prediction',\\\n",
    "            metricName='f1')\n",
    "f1_dt = evaluator.evaluate(result_dt) * 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[2440    0]\n",
      " [  15 2412]]\n"
     ]
    }
   ],
   "source": [
    "# Matriz de confusão\n",
    "y_true = result_dt.select(\"label\").toPandas()\n",
    "y_pred = result_dt.select(\"prediction\").toPandas()\n",
    "mc_dt = confusion_matrix(y_true, y_pred, labels = [0,1])\n",
    "tn_dt, fp_dt, fn_dt, tp_dt = confusion_matrix(y_true, y_pred).ravel()\n",
    "print(mc_dt)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "fp_dt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 99,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "4867"
      ]
     },
     "execution_count": 99,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.sum(mc_dt)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7f7e94afc470>"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "group_names = ['Verdadeiro Negativo','Falso Positivo','Falso Negativo','Verdadeiro Positivo']\n",
    "group_counts = ['{0:0.0f}'.format(value) for value in mc_dt.flatten()]\n",
    "group_percentages = ['{0:.2%}'.format(value) for value in mc_dt.flatten()/np.sum(mc_dt)]\n",
    "labels = [f'{v1}\\n{v2}\\n{v3}' for v1, v2, v3 in zip(group_names,group_counts,group_percentages)]\n",
    "labels = np.asarray(labels).reshape(2,2)\n",
    "sns.heatmap(mc_dt, annot=labels, fmt='', cmap='Blues')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Resultados do modelo Decision Tree:\n",
      "+--------+------+--------+--------+--------------+--------------+-----------------+--------------+\n",
      "|acurácia|recall|precisão|f1 score|falso positivo|falso negativo|tempo treinamento|tempo predição|\n",
      "+--------+------+--------+--------+--------------+--------------+-----------------+--------------+\n",
      "|   99.69| 99.69|   99.69|   99.69|             0|            15|             8.95|          0.08|\n",
      "+--------+------+--------+--------+--------------+--------------+-----------------+--------------+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# Exibição dos resultados\n",
    "evaluator_dt = spark.createDataFrame(\n",
    "    [(round(accuracy_dt,2), round(recall_dt,2), round(precision_dt,2), round(f1_dt,2),\\\n",
    "      int(fp_dt), int(fn_dt),\\\n",
    "      round(time_dt_train,2), round(time_dt_pred,2))],\\\n",
    "    ['acurácia','recall','precisão','f1 score',\\\n",
    "     'falso positivo', 'falso negativo',\\\n",
    "     'tempo treinamento','tempo predição'])\n",
    "print(\"Resultados do modelo Decision Tree:\")\n",
    "evaluator_dt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+-----+-------------------------------------+-------------+----------------------------------------+----------+\n",
      "|label|features                             |rawPrediction|probability                             |prediction|\n",
      "+-----+-------------------------------------+-------------+----------------------------------------+----------+\n",
      "|0.0  |(7,[0,1,4],[1.0,10782.94,100585.0])  |[1128.0,16.0]|[0.986013986013986,0.013986013986013986]|0.0       |\n",
      "|0.0  |(7,[0,1,4],[1.0,17851.54,32353.0])   |[1128.0,16.0]|[0.986013986013986,0.013986013986013986]|0.0       |\n",
      "|0.0  |(7,[0,1,4],[1.0,26004.52,52251.58])  |[1128.0,16.0]|[0.986013986013986,0.013986013986013986]|0.0       |\n",
      "|0.0  |(7,[0,1,4],[1.0,47913.58,51304.0])   |[1128.0,16.0]|[0.986013986013986,0.013986013986013986]|0.0       |\n",
      "|0.0  |(7,[0,1,4],[1.0,193605.38,249452.05])|[1128.0,16.0]|[0.986013986013986,0.013986013986013986]|0.0       |\n",
      "+-----+-------------------------------------+-------------+----------------------------------------+----------+\n",
      "only showing top 5 rows\n",
      "\n"
     ]
    }
   ],
   "source": [
    "result_dt.show(truncate = False, n = 5)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Random Forest (RF)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Treinamento do modelo de predição\n",
    "start_time = time.time()\n",
    "trainer = RandomForestClassifier(featuresCol='features', labelCol='label', predictionCol='prediction', probabilityCol='probability',\\\n",
    "                                 rawPredictionCol='rawPrediction', maxDepth=5, maxBins=32, minInstancesPerNode=1, minInfoGain=0.0,\\\n",
    "                                 numTrees=50, featureSubsetStrategy='auto', seed=None, subsamplingRate=1.0,\\\n",
    "                                 maxMemoryInMB=256, cacheNodeIds=False, checkpointInterval=10, impurity='gini')\n",
    "model_rf = trainer.fit(train)\n",
    "time_rf_train = time.time() - start_time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Execução do modelo de predição na base de teste\n",
    "start_time = time.time()\n",
    "result_rf = model_rf.transform(test)\n",
    "time_rf_pred = time.time() - start_time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Cálculo da acurácia do modelo de predição\n",
    "evaluator = MulticlassClassificationEvaluator(labelCol='label', predictionCol='prediction',\\\n",
    "            metricName='accuracy')\n",
    "accuracy_rf = evaluator.evaluate(result_rf) * 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Cálculo do recall do modelo de predição\n",
    "evaluator = MulticlassClassificationEvaluator(labelCol='label', predictionCol='prediction',\\\n",
    "            metricName='weightedRecall')\n",
    "recall_rf = evaluator.evaluate(result_rf) * 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Cálculo da precisão do modelo de predição\n",
    "evaluator = MulticlassClassificationEvaluator(labelCol='label', predictionCol='prediction',\\\n",
    "            metricName='weightedPrecision')\n",
    "precision_rf = evaluator.evaluate(result_rf) * 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Cálculo da F1 score do modelo de predição\n",
    "evaluator = MulticlassClassificationEvaluator(labelCol='label', predictionCol='prediction',\\\n",
    "            metricName='f1')\n",
    "f1_rf = evaluator.evaluate(result_rf) * 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[2440    0]\n",
      " [  16 2411]]\n"
     ]
    }
   ],
   "source": [
    "# Matriz de confusão\n",
    "y_true = result_rf.select(\"label\").toPandas()\n",
    "y_pred = result_rf.select(\"prediction\").toPandas()\n",
    "mc_rf = confusion_matrix(y_true, y_pred)\n",
    "tn_rf, fp_rf, fn_rf, tp_rf = confusion_matrix(y_true, y_pred).ravel()\n",
    "print(mc_rf)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7f7e943acb38>"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "group_names = ['Verdadeiro Negativo','Falso Positivo','Falso Negativo','Verdadeiro Positivo']\n",
    "group_counts = ['{0:0.0f}'.format(value) for value in mc_dt.flatten()]\n",
    "group_percentages = ['{0:.2%}'.format(value) for value in mc_rf.flatten()/np.sum(mc_rf)]\n",
    "labels = [f'{v1}\\n{v2}\\n{v3}' for v1, v2, v3 in zip(group_names,group_counts,group_percentages)]\n",
    "labels = np.asarray(labels).reshape(2,2)\n",
    "sns.heatmap(mc_rf, annot=labels, fmt='', cmap='Blues')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Resultados do modelo Random Forest:\n",
      "+--------+------+--------+--------+--------------+--------------+-----------------+--------------+\n",
      "|acurácia|recall|precisão|f1 score|falso positivo|falso negativo|tempo treinamento|tempo predição|\n",
      "+--------+------+--------+--------+--------------+--------------+-----------------+--------------+\n",
      "|   99.67| 99.67|   99.67|   99.67|             0|            16|            10.01|          0.09|\n",
      "+--------+------+--------+--------+--------------+--------------+-----------------+--------------+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# Exibição dos resultados\n",
    "evaluator_rf = spark.createDataFrame(\n",
    "    [(round(accuracy_rf,2), round(recall_rf,2), round(precision_rf,2), round(f1_rf,2),\\\n",
    "      int(fp_rf), int(fn_rf),\\\n",
    "      round(time_rf_train,2), round(time_rf_pred,2))],\\\n",
    "    ['acurácia','recall','precisão','f1 score',\\\n",
    "     'falso positivo', 'falso negativo',\\\n",
    "     'tempo treinamento','tempo predição'])\n",
    "print(\"Resultados do modelo Random Forest:\")\n",
    "evaluator_rf.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+-----+-------------------------------------+--------------------------------------+-----------------------------------------+----------+\n",
      "|label|features                             |rawPrediction                         |probability                              |prediction|\n",
      "+-----+-------------------------------------+--------------------------------------+-----------------------------------------+----------+\n",
      "|0.0  |(7,[0,1,4],[1.0,10782.94,100585.0])  |[46.884291017466246,3.11570898253376] |[0.9376858203493248,0.062314179650675185]|0.0       |\n",
      "|0.0  |(7,[0,1,4],[1.0,17851.54,32353.0])   |[46.55485970519186,3.445140294808147] |[0.9310971941038371,0.06890280589616293] |0.0       |\n",
      "|0.0  |(7,[0,1,4],[1.0,26004.52,52251.58])  |[47.0295287281215,2.970471271878508]  |[0.9405905745624298,0.05940942543757015] |0.0       |\n",
      "|0.0  |(7,[0,1,4],[1.0,47913.58,51304.0])   |[49.06861786549592,0.9313821345040785]|[0.9813723573099185,0.018627642690081568]|0.0       |\n",
      "|0.0  |(7,[0,1,4],[1.0,193605.38,249452.05])|[49.25780705468512,0.7421929453148892]|[0.9851561410937022,0.014843858906297783]|0.0       |\n",
      "+-----+-------------------------------------+--------------------------------------+-----------------------------------------+----------+\n",
      "only showing top 5 rows\n",
      "\n"
     ]
    }
   ],
   "source": [
    "result_rf.show(truncate = False, n = 5)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Neural Network Perceptron (NNP)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Treinamento do modelo de predição\n",
    "start_time =  time.time()\n",
    "layers = [7, 5, 5, 2]\n",
    "trainer = MultilayerPerceptronClassifier(featuresCol='features', labelCol='label',\\\n",
    "          maxIter=100, layers=layers, blockSize=128, seed=1234)\n",
    "model_nnp = trainer.fit(train)\n",
    "time_nnp_train = time.time() - start_time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Execução do modelo de predição na base de teste\n",
    "start_time =  time.time()\n",
    "result_nnp = model_nnp.transform(test)\n",
    "time_nnp_pred = time.time() - start_time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Cálculo da acurácia do modelo de predição\n",
    "evaluator = MulticlassClassificationEvaluator(labelCol='label', predictionCol='prediction',\\\n",
    "            metricName='accuracy')\n",
    "accuracy_nnp = evaluator.evaluate(result_nnp) * 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Cálculo do recall do modelo de predição\n",
    "evaluator = MulticlassClassificationEvaluator(labelCol='label', predictionCol='prediction',\\\n",
    "            metricName='weightedRecall')\n",
    "recall_nnp = evaluator.evaluate(result_nnp) * 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Cálculo da precisão do modelo de predição\n",
    "evaluator = MulticlassClassificationEvaluator(labelCol='label', predictionCol='prediction',\\\n",
    "            metricName='weightedPrecision')\n",
    "precision_nnp = evaluator.evaluate(result_nnp) * 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Cálculo da F1 score do modelo de predição\n",
    "evaluator = MulticlassClassificationEvaluator(labelCol='label', predictionCol='prediction',\\\n",
    "            metricName='f1')\n",
    "f1_nnp = evaluator.evaluate(result_nnp) * 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[2221  219]\n",
      " [  57 2370]]\n"
     ]
    }
   ],
   "source": [
    "# Matriz de confusão\n",
    "y_true = result_nnp.select(\"label\").toPandas()\n",
    "y_pred = result_nnp.select(\"prediction\").toPandas()\n",
    "mc_nnp = confusion_matrix(y_true, y_pred)\n",
    "tn_nnp, fp_nnp, fn_nnp, tp_nnp = confusion_matrix(y_true, y_pred).ravel()\n",
    "print(mc_nnp)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7f7e942f3fd0>"
      ]
     },
     "execution_count": 51,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "group_names = ['Verdadeiro Negativo','Falso Positivo','Falso Negativo','Verdadeiro Positivo']\n",
    "group_counts = ['{0:0.0f}'.format(value) for value in mc_dt.flatten()]\n",
    "group_percentages = ['{0:.2%}'.format(value) for value in mc_nnp.flatten()/np.sum(mc_nnp)]\n",
    "labels = [f'{v1}\\n{v2}\\n{v3}' for v1, v2, v3 in zip(group_names,group_counts,group_percentages)]\n",
    "labels = np.asarray(labels).reshape(2,2)\n",
    "sns.heatmap(mc_nnp, annot=labels, fmt='', cmap='Blues')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Resultados do modelo Neural Network Perceptron:\n",
      "+--------+------+--------+--------+--------------+--------------+-----------------+--------------+\n",
      "|acurácia|recall|precisão|f1 score|falso positivo|falso negativo|tempo treinamento|tempo predição|\n",
      "+--------+------+--------+--------+--------------+--------------+-----------------+--------------+\n",
      "|   94.33| 94.33|   94.53|   94.32|           219|            57|            53.09|          0.07|\n",
      "+--------+------+--------+--------+--------------+--------------+-----------------+--------------+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# Exibição dos resultados\n",
    "evaluator_nnp = spark.createDataFrame(\n",
    "    [(round(accuracy_nnp,2), round(recall_nnp,2), round(precision_nnp,2), round(f1_nnp,2),\\\n",
    "      int(fp_nnp), int(fn_nnp),\\\n",
    "      round(time_nnp_train,2), round(time_nnp_pred,2))],\\\n",
    "    ['acurácia','recall','precisão','f1 score',\\\n",
    "     'falso positivo', 'falso negativo',\\\n",
    "     'tempo treinamento','tempo predição'])\n",
    "print(\"Resultados do modelo Neural Network Perceptron:\")\n",
    "evaluator_nnp.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+-----+-------------------------------------+----------------------------------------+----------------------------------------+----------+\n",
      "|label|features                             |rawPrediction                           |probability                             |prediction|\n",
      "+-----+-------------------------------------+----------------------------------------+----------------------------------------+----------+\n",
      "|0.0  |(7,[0,1,4],[1.0,10782.94,100585.0])  |[0.7711767786867174,0.2048427240060645] |[0.637916847820783,0.362083152179217]   |0.0       |\n",
      "|0.0  |(7,[0,1,4],[1.0,17851.54,32353.0])   |[0.7711767786867174,0.2048427240060645] |[0.637916847820783,0.362083152179217]   |0.0       |\n",
      "|0.0  |(7,[0,1,4],[1.0,26004.52,52251.58])  |[0.7711767786867174,0.2048427240060645] |[0.637916847820783,0.362083152179217]   |0.0       |\n",
      "|0.0  |(7,[0,1,4],[1.0,47913.58,51304.0])   |[1.6647284287113397,-0.7801385349979125]|[0.9201852705430213,0.07981472945697869]|0.0       |\n",
      "|0.0  |(7,[0,1,4],[1.0,193605.38,249452.05])|[0.7711767786867174,0.2048427240060645] |[0.637916847820783,0.362083152179217]   |0.0       |\n",
      "+-----+-------------------------------------+----------------------------------------+----------------------------------------+----------+\n",
      "only showing top 5 rows\n",
      "\n"
     ]
    }
   ],
   "source": [
    "result_nnp.show(truncate = False, n = 5)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Naive Bayes (NB)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Treinamento do modelo de predição\n",
    "start_time =  time.time()\n",
    "trainer = NaiveBayes(smoothing=1.0, modelType='multinomial')\n",
    "model_nb = trainer.fit(train)\n",
    "time_nb_train = time.time() - start_time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Execução do modelo de predição na base de teste\n",
    "start_time =  time.time()\n",
    "result_nb = model_nb.transform(test)\n",
    "time_nb_pred = time.time() - start_time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Cálculo da acurácia do modelo de predição\n",
    "evaluator = MulticlassClassificationEvaluator(labelCol='label', predictionCol='prediction',\\\n",
    "            metricName='accuracy')\n",
    "accuracy_nb = evaluator.evaluate(result_nb) * 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Cálculo do recall do modelo de predição\n",
    "evaluator = MulticlassClassificationEvaluator(labelCol='label', predictionCol='prediction',\\\n",
    "            metricName='weightedRecall')\n",
    "recall_nb = evaluator.evaluate(result_nb) * 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Cálculo da precisão do modelo de predição\n",
    "evaluator = MulticlassClassificationEvaluator(labelCol='label', predictionCol='prediction',\\\n",
    "            metricName='weightedPrecision')\n",
    "precision_nb = evaluator.evaluate(result_nb) * 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Cálculo da F1 score do modelo de predição\n",
    "evaluator = MulticlassClassificationEvaluator(labelCol='label', predictionCol='prediction',\\\n",
    "            metricName='f1')\n",
    "f1_nb = evaluator.evaluate(result_nb) * 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[1500  940]\n",
      " [ 309 2118]]\n"
     ]
    }
   ],
   "source": [
    "# Matriz de confusão\n",
    "y_true = result_nb.select(\"label\").toPandas()\n",
    "y_pred = result_nb.select(\"prediction\").toPandas()\n",
    "mc_nb = confusion_matrix(y_true, y_pred)\n",
    "tn_nb, fp_nb, fn_nb, tp_nb = confusion_matrix(y_true, y_pred).ravel()\n",
    "print(mc_nb)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7f7e200f0d30>"
      ]
     },
     "execution_count": 61,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "group_names = ['Verdadeiro Negativo','Falso Positivo','Falso Negativo','Verdadeiro Positivo']\n",
    "group_counts = ['{0:0.0f}'.format(value) for value in mc_dt.flatten()]\n",
    "group_percentages = ['{0:.2%}'.format(value) for value in mc_nb.flatten()/np.sum(mc_nb)]\n",
    "labels = [f'{v1}\\n{v2}\\n{v3}' for v1, v2, v3 in zip(group_names,group_counts,group_percentages)]\n",
    "labels = np.asarray(labels).reshape(2,2)\n",
    "sns.heatmap(mc_nb, annot=labels, fmt='', cmap='Blues')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Resultados do modelo Naive Bayes:\n",
      "+--------+------+--------+--------+--------------+--------------+-----------------+--------------+\n",
      "|acurácia|recall|precisão|f1 score|falso positivo|falso negativo|tempo treinamento|tempo predição|\n",
      "+--------+------+--------+--------+--------------+--------------+-----------------+--------------+\n",
      "|   74.34| 74.34|   76.11|   73.91|           940|           309|             6.05|          0.06|\n",
      "+--------+------+--------+--------+--------------+--------------+-----------------+--------------+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# Exibição dos resultados\n",
    "evaluator_nb = spark.createDataFrame(\n",
    "    [(round(accuracy_nb,2), round(recall_nb,2), round(precision_nb,2), round(f1_nb,2),\\\n",
    "      int(fp_nb), int(fn_nb),\\\n",
    "      round(time_nb_train,2), round(time_nb_pred,2))],\\\n",
    "    ['acurácia','recall','precisão','f1 score',\\\n",
    "     'falso positivo', 'falso negativo',\\\n",
    "     'tempo treinamento','tempo predição'])\n",
    "print(\"Resultados do modelo Naive Bayes:\")\n",
    "evaluator_nb.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+-----+-------------------------------------+-----------------------------------------+-----------+----------+\n",
      "|label|features                             |rawPrediction                            |probability|prediction|\n",
      "+-----+-------------------------------------+-----------------------------------------+-----------+----------+\n",
      "|0.0  |(7,[0,1,4],[1.0,10782.94,100585.0])  |[-182269.67865277277,-241851.82351438346]|[1.0,0.0]  |0.0       |\n",
      "|0.0  |(7,[0,1,4],[1.0,17851.54,32353.0])   |[-112337.49642590462,-95769.90319682079] |[0.0,1.0]  |1.0       |\n",
      "|0.0  |(7,[0,1,4],[1.0,26004.52,52251.58])  |[-170868.12565460312,-151134.76757259917]|[0.0,1.0]  |1.0       |\n",
      "|0.0  |(7,[0,1,4],[1.0,47913.58,51304.0])   |[-251329.49680357386,-176358.5799792024] |[0.0,1.0]  |1.0       |\n",
      "|0.0  |(7,[0,1,4],[1.0,193605.38,249452.05])|[-1075006.6392050763,-808275.1143704863] |[0.0,1.0]  |1.0       |\n",
      "+-----+-------------------------------------+-----------------------------------------+-----------+----------+\n",
      "only showing top 5 rows\n",
      "\n"
     ]
    }
   ],
   "source": [
    "result_nb.show(truncate = False, n = 5)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Logistic Regression (LR)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Treinamento do modelo de predição\n",
    "start_time =  time.time()\n",
    "trainer = LogisticRegression(maxIter=10, tol=1E-6, fitIntercept=True)\n",
    "model_lr = trainer.fit(train)\n",
    "time_lr_train = time.time() - start_time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Execução do modelo de predição na base de teste\n",
    "start_time =  time.time()\n",
    "result_lr = model_lr.transform(test)\n",
    "time_lr_pred = time.time() - start_time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Cálculo da acurácia do modelo de predição\n",
    "evaluator = MulticlassClassificationEvaluator(labelCol='label', predictionCol='prediction',\\\n",
    "            metricName='accuracy')\n",
    "accuracy_lr = evaluator.evaluate(result_lr) * 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Cálculo do recall do modelo de predição\n",
    "evaluator = MulticlassClassificationEvaluator(labelCol='label', predictionCol='prediction',\\\n",
    "            metricName='weightedRecall')\n",
    "recall_lr = evaluator.evaluate(result_lr) * 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Cálculo da precisão do modelo de predição\n",
    "evaluator = MulticlassClassificationEvaluator(labelCol='label', predictionCol='prediction',\\\n",
    "            metricName='weightedPrecision')\n",
    "precision_lr = evaluator.evaluate(result_lr) * 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Cálculo da F1 score do modelo de predição\n",
    "evaluator = MulticlassClassificationEvaluator(labelCol='label', predictionCol='prediction',\\\n",
    "            metricName='f1')\n",
    "f1_lr = evaluator.evaluate(result_lr) * 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[2437    3]\n",
      " [  64 2363]]\n"
     ]
    }
   ],
   "source": [
    "# Matriz de confusão\n",
    "y_true = result_lr.select(\"label\").toPandas()\n",
    "y_pred = result_lr.select(\"prediction\").toPandas()\n",
    "mc_lr = confusion_matrix(y_true, y_pred)\n",
    "tn_lr, fp_lr, fn_lr, tp_lr = confusion_matrix(y_true, y_pred).ravel()\n",
    "print(mc_lr)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7f7e200f0be0>"
      ]
     },
     "execution_count": 71,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "group_names = ['Verdadeiro Negativo','Falso Positivo','Falso Negativo','Verdadeiro Positivo']\n",
    "group_counts = ['{0:0.0f}'.format(value) for value in mc_dt.flatten()]\n",
    "group_percentages = ['{0:.2%}'.format(value) for value in mc_lr.flatten()/np.sum(mc_lr)]\n",
    "labels = [f'{v1}\\n{v2}\\n{v3}' for v1, v2, v3 in zip(group_names,group_counts,group_percentages)]\n",
    "labels = np.asarray(labels).reshape(2,2)\n",
    "sns.heatmap(mc_lr, annot=labels, fmt='', cmap='Blues')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Resultados do modelo Logistic Regression:\n",
      "+--------+------+--------+--------+--------------+--------------+-----------------+--------------+\n",
      "|acurácia|recall|precisão|f1 score|falso positivo|falso negativo|tempo treinamento|tempo predição|\n",
      "+--------+------+--------+--------+--------------+--------------+-----------------+--------------+\n",
      "|   98.62| 98.62|   98.65|   98.62|             3|            64|             8.81|          0.02|\n",
      "+--------+------+--------+--------+--------------+--------------+-----------------+--------------+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# Exibição dos resultados\n",
    "evaluator_lr = spark.createDataFrame(\n",
    "    [(round(accuracy_lr,2), round(recall_lr,2), round(precision_lr,2), round(f1_lr,2),\\\n",
    "      int(fp_lr), int(fn_lr),\\\n",
    "      round(time_lr_train,2), round(time_lr_pred,2))],\\\n",
    "    ['acurácia','recall','precisão','f1 score',\\\n",
    "     'falso positivo', 'falso negativo',\\\n",
    "     'tempo treinamento','tempo predição'])\n",
    "print(\"Resultados do modelo Logistic Regression:\")\n",
    "evaluator_lr.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+-----+-------------------------------------+----------------------------------------+----------------------------------------+----------+\n",
      "|label|features                             |rawPrediction                           |probability                             |prediction|\n",
      "+-----+-------------------------------------+----------------------------------------+----------------------------------------+----------+\n",
      "|0.0  |(7,[0,1,4],[1.0,10782.94,100585.0])  |[1.6632349203838523,-1.6632349203838523]|[0.8406717748203733,0.15932822517962675]|0.0       |\n",
      "|0.0  |(7,[0,1,4],[1.0,17851.54,32353.0])   |[1.6721632731207916,-1.6721632731207916]|[0.8418640285420456,0.15813597145795436]|0.0       |\n",
      "|0.0  |(7,[0,1,4],[1.0,26004.52,52251.58])  |[1.641626350124325,-1.641626350124325]  |[0.8377561136872689,0.1622438863127311] |0.0       |\n",
      "|0.0  |(7,[0,1,4],[1.0,47913.58,51304.0])   |[1.5821045075587057,-1.5821045075587057]|[0.829502361107439,0.17049763889256095] |0.0       |\n",
      "|0.0  |(7,[0,1,4],[1.0,193605.38,249452.05])|[1.101619638406227,-1.101619638406227]  |[0.7505634540254045,0.24943654597459555]|0.0       |\n",
      "+-----+-------------------------------------+----------------------------------------+----------------------------------------+----------+\n",
      "only showing top 5 rows\n",
      "\n"
     ]
    }
   ],
   "source": [
    "result_lr.show(truncate = False, n = 5)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Suport Vector Machines (SVM)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Treinamento do modelo de predição\n",
    "start_time =  time.time()\n",
    "trainer = LinearSVC(featuresCol='features', labelCol='label',\\\n",
    "                    maxIter=100, regParam=0.1)\n",
    "model_svm = trainer.fit(train)\n",
    "time_svm_train = time.time() - start_time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Execução do modelo de predição na base de teste\n",
    "start_time =  time.time()\n",
    "result_svm = model_svm.transform(test)\n",
    "time_svm_pred = time.time() - start_time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Cálculo da acurácia do modelo de predição\n",
    "evaluator = MulticlassClassificationEvaluator(labelCol='label', predictionCol='prediction',\\\n",
    "            metricName='accuracy')\n",
    "accuracy_svm = evaluator.evaluate(result_svm) * 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 77,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Cálculo do recall do modelo de predição\n",
    "evaluator = MulticlassClassificationEvaluator(labelCol='label', predictionCol='prediction',\\\n",
    "            metricName='weightedRecall')\n",
    "recall_svm = evaluator.evaluate(result_svm) * 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Cálculo da precisão do modelo de predição\n",
    "evaluator = MulticlassClassificationEvaluator(labelCol='label', predictionCol='prediction',\\\n",
    "            metricName='weightedPrecision')\n",
    "precision_svm = evaluator.evaluate(result_svm) * 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 79,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Cálculo da F1 score do modelo de predição\n",
    "evaluator = MulticlassClassificationEvaluator(labelCol='label', predictionCol='prediction',\\\n",
    "            metricName='f1')\n",
    "f1_svm = evaluator.evaluate(result_svm) * 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 80,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[2209  231]\n",
      " [ 123 2304]]\n"
     ]
    }
   ],
   "source": [
    "# Matriz de confusão\n",
    "y_true = result_svm.select(\"label\").toPandas()\n",
    "y_pred = result_svm.select(\"prediction\").toPandas()\n",
    "mc_svm = confusion_matrix(y_true, y_pred)\n",
    "tn_svm, fp_svm, fn_svm, tp_svm = confusion_matrix(y_true, y_pred).ravel()\n",
    "print(mc_svm)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 81,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7f7e94b04940>"
      ]
     },
     "execution_count": 81,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "group_names = ['Verdadeiro Negativo','Falso Positivo','Falso Negativo','Verdadeiro Positivo']\n",
    "group_counts = ['{0:0.0f}'.format(value) for value in mc_dt.flatten()]\n",
    "group_percentages = ['{0:.2%}'.format(value) for value in mc_svm.flatten()/np.sum(mc_svm)]\n",
    "labels = [f'{v1}\\n{v2}\\n{v3}' for v1, v2, v3 in zip(group_names,group_counts,group_percentages)]\n",
    "labels = np.asarray(labels).reshape(2,2)\n",
    "sns.heatmap(mc_svm, annot=labels, fmt='', cmap='Blues')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 82,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Resultados do modelo Suport Vector Machines:\n",
      "+--------+------+--------+--------+--------------+--------------+-----------------+--------------+\n",
      "|acurácia|recall|precisão|f1 score|falso positivo|falso negativo|tempo treinamento|tempo predição|\n",
      "+--------+------+--------+--------+--------------+--------------+-----------------+--------------+\n",
      "|   92.73| 92.73|   92.81|   92.72|           231|           123|           585.06|          0.08|\n",
      "+--------+------+--------+--------+--------------+--------------+-----------------+--------------+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# Exibição dos resultados\n",
    "evaluator_svm = spark.createDataFrame(\n",
    "    [(round(accuracy_svm,2), round(recall_svm,2), round(precision_svm,2), round(f1_svm,2),\\\n",
    "      int(fp_svm), int(fn_svm),\\\n",
    "      round(time_svm_train,2), round(time_svm_pred,2))],\\\n",
    "    ['acurácia','recall','precisão','f1 score',\\\n",
    "     'falso positivo', 'falso negativo',\\\n",
    "     'tempo treinamento','tempo predição'])\n",
    "print(\"Resultados do modelo Suport Vector Machines:\")\n",
    "evaluator_svm.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+-----+-------------------------------------+------------------------------------------+----------+\n",
      "|label|features                             |rawPrediction                             |prediction|\n",
      "+-----+-------------------------------------+------------------------------------------+----------+\n",
      "|0.0  |(7,[0,1,4],[1.0,10782.94,100585.0])  |[-0.18166026181359599,0.18166026181359599]|1.0       |\n",
      "|0.0  |(7,[0,1,4],[1.0,17851.54,32353.0])   |[-0.1860012133919604,0.1860012133919604]  |1.0       |\n",
      "|0.0  |(7,[0,1,4],[1.0,26004.52,52251.58])  |[-0.1866648694328334,0.1866648694328334]  |1.0       |\n",
      "|0.0  |(7,[0,1,4],[1.0,47913.58,51304.0])   |[-0.19084547190768078,0.19084547190768078]|1.0       |\n",
      "|0.0  |(7,[0,1,4],[1.0,193605.38,249452.05])|[-0.2096398403889629,0.2096398403889629]  |1.0       |\n",
      "+-----+-------------------------------------+------------------------------------------+----------+\n",
      "only showing top 5 rows\n",
      "\n"
     ]
    }
   ],
   "source": [
    "result_svm.show(truncate = False, n = 5)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Resultados"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 84,
   "metadata": {},
   "outputs": [],
   "source": [
    "models = ['Decision Tree','Random Forest','Neural Network Perceptron','Naive Bayes','Logistic Regression','Suport Vector Machines']"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Accuracy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 85,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Ranking Accuracy %\n",
      "+-------------------------+-----------------+\n",
      "|Modelo                   |Acuracia         |\n",
      "+-------------------------+-----------------+\n",
      "|Decision Tree            |99.69180193137457|\n",
      "|Random Forest            |99.6712553934662 |\n",
      "|Logistic Regression      |98.62338196013971|\n",
      "|Neural Network Perceptron|94.32915553729197|\n",
      "|Naive Bayes              |74.33737415245531|\n",
      "|Suport Vector Machines   |74.33737415245531|\n",
      "+-------------------------+-----------------+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "print('Ranking Accuracy %')\n",
    "\n",
    "list = ((models[0],accuracy_dt),\\\n",
    "        (models[1],accuracy_rf),\\\n",
    "        (models[2],accuracy_nnp),\\\n",
    "        (models[3],accuracy_nb),\\\n",
    "        (models[4],accuracy_lr),\\\n",
    "        (models[5],accuracy_nb))\n",
    "df_acuracia = spark.createDataFrame(list, ['Modelo', 'Acuracia'])\n",
    "df_acuracia.sort(df_acuracia.Acuracia.desc()).show(truncate = False)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Recall"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 86,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Ranking\n",
      "+-------------------------+-----------------+\n",
      "|Modelo                   |Recall           |\n",
      "+-------------------------+-----------------+\n",
      "|Decision Tree            |99.69180193137457|\n",
      "|Random Forest            |99.6712553934662 |\n",
      "|Logistic Regression      |98.62338196013971|\n",
      "|Neural Network Perceptron|94.32915553729198|\n",
      "|Suport Vector Machines   |92.7265255804397 |\n",
      "|Naive Bayes              |74.33737415245531|\n",
      "+-------------------------+-----------------+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "print('Ranking')\n",
    "\n",
    "list = ((models[0],recall_dt),\\\n",
    "        (models[1],recall_rf),\\\n",
    "        (models[2],recall_nnp),\\\n",
    "        (models[3],recall_nb),\\\n",
    "        (models[4],recall_lr),\\\n",
    "        (models[5],recall_svm))\n",
    "df_recall = spark.createDataFrame(list, ['Modelo', 'Recall'])\n",
    "df_recall.sort(df_recall.Recall.desc()).show(truncate = False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Precision"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 87,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Ranking Precision %\n",
      "+-------------------------+-----------------+\n",
      "|Modelo                   |Precisao         |\n",
      "+-------------------------+-----------------+\n",
      "|Decision Tree            |99.69368501529692|\n",
      "|Random Forest            |99.67339705214069|\n",
      "|Logistic Regression      |98.65386541768162|\n",
      "|Neural Network Perceptron|94.52742561372798|\n",
      "|Suport Vector Machines   |92.8116914357235 |\n",
      "|Naive Bayes              |76.1080875663157 |\n",
      "+-------------------------+-----------------+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "print('Ranking Precision %')\n",
    "\n",
    "list = ((models[0],precision_dt),\\\n",
    "        (models[1],precision_rf),\\\n",
    "        (models[2],precision_nnp),\\\n",
    "        (models[3],precision_nb),\\\n",
    "        (models[4],precision_lr),\\\n",
    "        (models[5],precision_svm))\n",
    "df_precision = spark.createDataFrame(list, ['Modelo', 'Precisao'])\n",
    "df_precision.sort(df_precision.Precisao.desc()).show(truncate = False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### F1 score"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 88,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+-------------------------+-----------------+\n",
      "|Modelo                   |F1               |\n",
      "+-------------------------+-----------------+\n",
      "|Decision Tree            |99.69179646661719|\n",
      "|Random Forest            |99.67124895371123|\n",
      "|Logistic Regression      |98.62311956684768|\n",
      "|Neural Network Perceptron|94.32337146710331|\n",
      "|Suport Vector Machines   |92.72337397325676|\n",
      "|Naive Bayes              |73.907980556053  |\n",
      "+-------------------------+-----------------+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "list = ((models[0],f1_dt),\\\n",
    "        (models[1],f1_rf),\\\n",
    "        (models[2],f1_nnp),\\\n",
    "        (models[3],f1_nb),\\\n",
    "        (models[4],f1_lr),\\\n",
    "        (models[5],f1_svm))\n",
    "df_f1 = spark.createDataFrame(list, ['Modelo', 'F1'])\n",
    "df_f1.sort(df_f1.F1.desc()).show(truncate = False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Tempo de Treinamento"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 89,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+-------------------------+------------------+\n",
      "|Modelo                   |Tempo_Treinamento |\n",
      "+-------------------------+------------------+\n",
      "|Naive Bayes              |6.047499179840088 |\n",
      "|Logistic Regression      |8.806694507598877 |\n",
      "|Decision Tree            |8.950883865356445 |\n",
      "|Random Forest            |10.0124351978302  |\n",
      "|Neural Network Perceptron|53.085105895996094|\n",
      "|Suport Vector Machines   |585.063728094101  |\n",
      "+-------------------------+------------------+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "list = ((models[0],time_dt_train),\\\n",
    "        (models[1],time_rf_train),\\\n",
    "        (models[2],time_nnp_train),\\\n",
    "        (models[3],time_nb_train),\\\n",
    "        (models[4],time_lr_train),\\\n",
    "        (models[5],time_svm_train))\n",
    "df_time_train = spark.createDataFrame(list, ['Modelo', 'Tempo_Treinamento'])\n",
    "df_time_train.sort(df_time_train.Tempo_Treinamento.asc()).show(truncate = False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Tempo de Predição"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 90,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+-------------------------+-------------------+\n",
      "|Modelo                   |Tempo_Predicao     |\n",
      "+-------------------------+-------------------+\n",
      "|Neural Network Perceptron|0.0682060718536377 |\n",
      "|Naive Bayes              |0.0682060718536377 |\n",
      "|Logistic Regression      |0.0682060718536377 |\n",
      "|Suport Vector Machines   |0.0682060718536377 |\n",
      "|Decision Tree            |0.08316850662231445|\n",
      "|Random Forest            |0.0926060676574707 |\n",
      "+-------------------------+-------------------+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "list = ((models[0],time_dt_pred),\\\n",
    "        (models[1],time_rf_pred),\\\n",
    "        (models[2],time_nnp_pred),\\\n",
    "        (models[3],time_nnp_pred),\\\n",
    "        (models[4],time_nnp_pred),\\\n",
    "        (models[5],time_nnp_pred))\n",
    "df_time_pred = spark.createDataFrame(list, ['Modelo', 'Tempo_Predicao'])\n",
    "df_time_pred.sort(df_time_pred.Tempo_Predicao.asc()).show(truncate = False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Falso Positivo"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 91,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+-------------------------+--------------+\n",
      "|Modelo                   |Falso_Positivo|\n",
      "+-------------------------+--------------+\n",
      "|Random Forest            |0             |\n",
      "|Decision Tree            |0             |\n",
      "|Logistic Regression      |3             |\n",
      "|Neural Network Perceptron|219           |\n",
      "|Suport Vector Machines   |231           |\n",
      "|Naive Bayes              |940           |\n",
      "+-------------------------+--------------+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "list = ((models[0],int(fp_dt)),\\\n",
    "        (models[1],int(fp_rf)),\\\n",
    "        (models[2],int(fp_nnp)),\\\n",
    "        (models[3],int(fp_nb)),\\\n",
    "        (models[4],int(fp_lr)),\\\n",
    "        (models[5],int(fp_svm)))\n",
    "df_fp = spark.createDataFrame(list, ['Modelo', 'Falso_Positivo'])\n",
    "df_fp.sort(df_fp.Falso_Positivo.asc()).show(truncate = False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Falso Negativo"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 92,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+-------------------------+--------------+\n",
      "|Modelo                   |Falso_Negativo|\n",
      "+-------------------------+--------------+\n",
      "|Decision Tree            |15            |\n",
      "|Random Forest            |16            |\n",
      "|Neural Network Perceptron|57            |\n",
      "|Logistic Regression      |64            |\n",
      "|Suport Vector Machines   |123           |\n",
      "|Naive Bayes              |309           |\n",
      "+-------------------------+--------------+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "list = ((models[0],int(fn_dt)),\\\n",
    "        (models[1],int(fn_rf)),\\\n",
    "        (models[2],int(fn_nnp)),\\\n",
    "        (models[3],int(fn_nb)),\\\n",
    "        (models[4],int(fn_lr)),\\\n",
    "        (models[5], int(fn_svm)))\n",
    "df_fn = spark.createDataFrame(list, ['Modelo', 'Falso_Negativo'])\n",
    "df_fn.sort(df_fn.Falso_Negativo.asc()).show(truncate = False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Comparativo"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 93,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+-------------------------+-----------------+--------------+--------------+------------------+-------------------+\n",
      "|Modelo                   |Acuracia         |Falso_Positivo|Falso_Negativo|Tempo_Treinamento |Tempo_Predicao     |\n",
      "+-------------------------+-----------------+--------------+--------------+------------------+-------------------+\n",
      "|Decision Tree            |99.69180193137457|0             |15            |8.950883865356445 |0.08316850662231445|\n",
      "|Random Forest            |99.6712553934662 |0             |16            |10.0124351978302  |0.0926060676574707 |\n",
      "|Logistic Regression      |98.62338196013971|3             |64            |8.806694507598877 |0.0682060718536377 |\n",
      "|Neural Network Perceptron|94.32915553729197|219           |57            |53.085105895996094|0.0682060718536377 |\n",
      "|Naive Bayes              |74.33737415245531|940           |309           |6.047499179840088 |0.0682060718536377 |\n",
      "|Suport Vector Machines   |74.33737415245531|231           |123           |585.063728094101  |0.0682060718536377 |\n",
      "+-------------------------+-----------------+--------------+--------------+------------------+-------------------+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "df = df_acuracia.join(df_fp, \"Modelo\")\n",
    "df = df.join(df_fn, \"Modelo\")\n",
    "df = df.join(df_time_train, \"Modelo\")\n",
    "df = df.join(df_time_pred, \"Modelo\")\n",
    "df.sort(df.Acuracia.desc()).show(truncate = False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Teste de classificação do melhor modelo"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+----+-------+-------------+--------------+--------------+--------------+---------+-----+\n",
      "|step| amount|oldbalanceOrg|newbalanceOrig|oldbalanceDest|newbalanceDest|indexType|label|\n",
      "+----+-------+-------------+--------------+--------------+--------------+---------+-----+\n",
      "|   1|1864.28|      21249.0|      19384.72|           0.0|           0.0|      2.0|  0.0|\n",
      "+----+-------+-------------+--------------+--------------+--------------+---------+-----+\n",
      "only showing top 1 row\n",
      "\n"
     ]
    }
   ],
   "source": [
    "df_selecionado.show(1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "metadata": {},
   "outputs": [],
   "source": [
    "#valores de entrada e saída para teste\n",
    "entradas = [1.0, 9839.64,170136.0, 160296.36,0.0,0.0,2.0]\n",
    "resultado = 0.0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 96,
   "metadata": {},
   "outputs": [],
   "source": [
    "#criando o dataframe Spark para execução do modelo\n",
    "df_teste = spark.createDataFrame(\n",
    "    [(resultado,\\\n",
    "      Vectors.dense([entradas[0],entradas[1],entradas[2],entradas[3],entradas[4],entradas[5],entradas[6]]))],\n",
    "    ['label', 'features'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 97,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+-----+----------+\n",
      "|label|prediction|\n",
      "+-----+----------+\n",
      "|  0.0|       0.0|\n",
      "+-----+----------+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "#resultado da predição utilizando o modelo RF\n",
    "resultado = model_rf.transform(df_teste)\n",
    "resultado.select('label','prediction').show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 98,
   "metadata": {},
   "outputs": [],
   "source": [
    "spark.stop()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}