breast_cancer_eda.Rmd

---
title: "Breast Cancer data analysis"
author: "Amazon Web Services"
date: "9/7/2021"
output: html_document
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
library(mlbench)
library(ggplot2)
library(caret)
```

## Breast Cancer data summary

This is an exploratory analysis on [UCI Breast Cancer Wisconsin (Diagnostic) dataset](https://archive.ics.uci.edu/ml/datasets/Breast+Cancer+Wisconsin+(Diagnostic)) from [mlbench](https://cran.r-project.org/web/packages/mlbench/index.html) library. The data is collected from 699 people who were eligible of the study. 9 features are computed from a digitized image of a fine needle aspirate (FNA) of a breast mass, describing the characteristics of the cell nuclei present in the image. Let's look at the descriptive statistics of the dataset that are in numeric format.

```{r breastcancer}
data(BreastCancer)
df <- BreastCancer
# convert input columns 2 to 10 from factor to numeric
for(i in 2:10) {
  df[,i] <- as.numeric(as.character(df[,i]))
}
summary(df)
```

## Histogram of clump thickness by class

We are interested to see the distribution of the clump thickness between the two classes: *Benign* and *Malignant*. 

```{r cl_thickness, echo=FALSE}
ggplot(df, aes(x=Cl.thickness))+
       geom_histogram(color="black", fill="white", binwidth = 1)+
       facet_grid(Class ~ .)
```

It turns out that *benign* cases tend to have smaller clumps as oppose to *malignant* cases who tend to have thicker clumps in the breasts.

## Training a machine learning model
Let's split the data, standardize accordingly and train a ML model. The training process includes a 10-fold cross validation using gradient boosting model, optimized with area under ROC curve.
```{r modeling}
# split the data into train and test and perform preprocessing
trainIndex <- createDataPartition(df$Class, p = .8, 
                                  list = FALSE, 
                                  times = 1)
df_train <- df[ trainIndex,]
df_test  <- df[-trainIndex,]
preProcValues <- preProcess(df_train, method = c("center", "scale", "medianImpute"))
df_train_transformed <- predict(preProcValues, df_train)

# train a model on df_train
fitControl <- trainControl(## 10-fold CV
  method = "repeatedcv",
  number = 10,
  ## repeated ten times
  repeats = 10,
  ## Estimate class probabilities
  classProbs = TRUE,
  ## Evaluate performance using 
  ## the following function
  summaryFunction = twoClassSummary)

set.seed(825)
gbmFit <- train(Class ~ ., data = df_train_transformed[,2:11], 
                method = "gbm", 
                trControl = fitControl,
                ## This last option is actually one
                ## for gbm() that passes through
                verbose = FALSE,
                metric = "ROC")
```

We can see the feature importance based on the algorithm.
```{r featureimportance, echo=FALSE}
summary(gbmFit)
```

This is the end of a simple analysis and plotting in a R Markdown file. We develop it in RStudio Workbench in Amazon SageMaker and will publish it to a RStudio Connect server.