BetikuOluwatobi/ibm-hydropower-challenge
Domain:
environment and energyRecord type:
project
Creator:
Bet
Host:
This solution placed 100th out of 444 with a public leaderboard RMSE of 6.8080 and a private leaderboard RMSE of 5.6184 for the Zindi Africa IBM SkillsBuild Challenge 2025
---
# IBM SkillsBuild Micro-Hydro Load Forecasting
Forecasting daily energy consumption (kWh) for off-grid micro-hydropower (MHP) systems using climate and operational data.
Developed for the Zindi Africa IBM SkillsBuild Challenge 2025, this solution placed **100th out of 444** with a public leaderboard RMSE of **6.8080** and a private leaderboard RMSE of **5.6184**.
## Requirements
- Python 3.11+
- ~8GB RAM recommended (higher if running full forecast mode(Note: Full forecast takes approximately 8 hours, to run lower forecast change windows in config.py to 4))
- Libraries in `requirements.txt`
- Data files from Zindi competition page
## Repo Structure
```
├── config.py # Global configuration paths and constants
├── data/ # Raw and processed data
│ ├── Climate Data/
│ │ ├── Kalam Climate Data.xlsx
│ └── Data/
│ └── Data.csv
├── modelling/
│ ├── model.py # Main LightGBM model pipeline and training logic
├── src/
│ ├── preprocessing.py # Data cleaning and preprocessing steps
│ ├── aggregation.py # Feature engineering and rolling window aggregations
│ ├── forecast.py # Prediction generation logic
├── results/ # Prediction outputs and intermediate files
├── run.py # Main pipeline entrypoint (train + predict)
├── requirements.txt
└── README.md
```
## Usage
### 1. Setup Environment
```bash
python -m venv venv
Linux: source venv/bin/activate, Windows: venv\Scripts\activate
pip install -r requirements.txt
```
### 2. Prepare Data
- Place the downloaded `Data.csv`, `SampleSubmission.csv`, and `Kalam Climate Data.xlsx` in their respective subdirectories under `data/`.
- Configure `config.py` if needed to reflect updated paths.
### 3. Run Forecast Pipeline
```bash
python run.py
```
This script:
- Loads and merges climate and MHP operational data.
- Performs preprocessing and rolling window feature engineering.
- Trains LightGBM models using k-fold …