DubaKanu/Milan-Traffic-Forecasting

Comparative Time Series Analysis and Forecasting of Mobile Network Traffic - African Leadership University # Milan Traffic Forecasting ## Comparative Time Series Analysis and Forecasting of Mobile Network Traffic **Author:** Josephine Duba Kanu **Institution:** African Leadership University **Course:** Machine Learning Techniques - I **Date:** May 2026 --- ## Project Overview This project analyzes and forecasts real-world mobile internet traffic data from the city of Milan, Italy, released by Telecom Italia Mobile (TIM) as part of their Big Data Challenge. The dataset covers 10,000 geographical areas over two months, with traffic recorded every 10 minutes. --- ## Models Implemented | Model | Type | MAE (Area 5181) | |-------|------|----------------| | Holt-Winters | Classical Statistical | 16.19 | | LSTM | Neural Network | 6.49 | | GRU | Neural Network | 7.05 | **Best Model: LSTM** — achieved lowest MAE and RMSE across majority of areas. --- ## Repository Structure Milan-Traffic-Forecasting/ ├── Milan_Traffic_Forecasting.ipynb ← Main notebook (all code) ├── README.md ← This file ├── requirements.txt ← Required libraries ├── area_5181.csv ← Highest traffic area data ├── area_4159.csv ← Area 4159 time series data ├── area_4556.csv ← Area 4556 time series data ├── task2_pdf.png ← PDF plot of traffic distribution ├── task2_timeseries.png ← Time series plots ├── task2_stationarity.png ← Stationarity analysis plot ├── task2_decomposition.png ← Decomposition plot ├── task2_acf_pacf.png ← ACF and PACF plots ├── task2_heatmap.png ← Spatial heatmap of Milan ├── task3_hw_predictions.png ← Holt-Winters predictions ├── task3_lstm_predictions.png ← LSTM predictions └── task3_gru_predictions.png ← GRU predictions --- ## How to Run the Code ### Requirements First install all required libraries: pip install -r requirements.txt Or install manually: pip install pand …