Sebaga-M/Botswana-inflation-forecast

# Botswana Food Price Inflation Forecast **Deep Learning IndabaX Botswana 2026 — Hackathon Submission** ## Project Overview This project forecasts Botswana's monthly food price inflation (% year-over-year) for January–December 2024, using historical data (2000–2023) across five datasets: global shipping costs (Baltic Dry Index), Brent crude oil prices, Botswana's central bank policy rate, Botswana's food/consumer prices, and cross-country inflation data for four regional trading partners. Two models were built and compared, as required: - **Classical baseline:** LightGBM (Gradient Boosted Trees) - **Deep learning model:** LSTM (Long Short-Term Memory neural network) LightGBM was selected as our best-performing model after honest evaluation — see `Model_Comparison_Report.pdf` for full analysis of both models, including the reasoning behind this choice. ## Repository Structure ├── data/ # Raw input datasets (5 CSV files) ├── notebooks/ │ └── full_analysis.ipynb # Complete analysis notebook (data prep, both models, forecast generation) ├── src/ │ ├── classical_model.py # LightGBM implementation │ └── deep_learning_model.py # LSTM implementation ├── outputs/ │ └── predictions_2024_v3.csv # Final submitted forecast ├── requirements.txt # Python dependencies └── README.md ## Setup Instructions 1. Clone this repository: 2. 2. Install dependencies: This builds sliding-window sequences, trains the LSTM, and outputs validation/test RMSE. **Full analysis (recommended for review):** Open `notebooks/full_analysis.ipynb` in Jupyter or Google Colab to see the complete step-by-step process: data merging, feature engineering, both models, data leak identification and correction, residual diagnostics, and final forecast generation. ## Key Results | Model | Test RMSE | Test MAE | |---|---|---| | **LightGBM (winner)** | **2.631** | **--** | | LSTM | 7.014 | 5.819 | | Naive baseline | 8.653 | 7.844 | LightGBM outperformed both the LSTM and a naive seasonal baseline. See ` …