Time Series Prediction of Internet Traffic

This project focuses on predicting future internet traffic using Call Detail Records (CDRs) provided by the Semantics and Knowledge Innovation Lab (SKIL). Utilizing PyTorch for deep learning and a dataset comprising two months of CDR data.

Dataset: https://www.nature.com/articles/sdata201555"

Project Overview

The goal of this project is to leverage time series analysis techniques to predict internet traffic accurately. This involves extensive data preprocessing and cleaning of the CDR dataset to make it suitable for time series forecasting. The final dataset is structured into grouped pickle sets for efficient training and testing.

Implemented models

• LSTM
• CONV 1D
• CONV 2D
• Deeper conv2D

Features

• Data Preprocessing: Cleaning and grouping CDR data into structured pickle format.
• Model Training: Using two months of CDR data with time intervals of every ten minutes.
• Prediction: Forecasting future internet traffic with the trained model.

Getting Started

Prerequisites

• Python 3.9
• Jupyter Notebook
• PyTorch

Acknowledgments

• Semantics and Knowledge Innovation Lab (SKIL) for providing the CDR dataset.
• PyTorch community for the deep learning framework.