womarianne/dtt-pathloss-benin

# DTT Path Loss Prediction in Benin — Reproducibility Repository Code and data accompanying the article on machine-learning-based path loss / field-strength prediction for Digital Terrestrial Television (DTT) coverage in Benin, comparing ML models against empirical models (Okumura-Hata, ITU-R) across two contrasting environments: **Cotonou** (urban) and **Kandi** (suburban). This repository contains only the notebooks, data, and precomputed results that were used to produce the figures and tables in the article — not the full thesis codebase (geospatial feature engineering, synthetic-data generation pipeline, etc.), which is kept separately. ## Repository structure ``` . ├── data/ │ ├── dataset2_Efield_clean.csv # 338 cleaned field measurements, ML-ready features │ └── radio_features_public.csv # Same measurements + zone/antenna-height metadata ├── notebooks/ # Run in numeric order — see below │ ├── 01_hyperparameter_optimization.ipynb │ ├── 02_feature_importance.ipynb │ ├── 03_hata_calibration.ipynb │ ├── 04_hybrid_models.ipynb │ ├── 05_augmentation_comparison.ipynb │ └── 06_cross_family_comparison.ipynb └── outputs/ # Precomputed CSV/JSON results (already saved by each notebook) ``` ## Setup Requires **Python 3.11+**. ```bash git clone github.com /dtt-pathloss-benin.git cd dtt-pathloss-benin pip install -r requirements.txt ``` ## Reproducing the results The notebooks must be run **in order** — each one loads results saved by the previous ones from `outputs/`. All `outputs/*.csv` / `*.json` files are already included, so you can either: - inspect the precomputed results directly, or - re-run any notebook end-to-end (it will overwrite its own outputs). | # | Notebook | What it does | Produces | |---|----------|---------------|----------| | 1 | `01_hyperparameter_optimization.ipynb` | Optuna hyperparameter search (RF, XGBoost, CatBoost, LightGBM) | `best_params.json`, …