Bonifacethuo/Ghana-s-Indigenous-Intel-Challenge-BEGINNERS-ONLY-Zindi

# 🌧️ RAIL: Indigenous Weather Forecasting - Complete Classification Pipeline This repository contains a comprehensive machine learning pipeline for predicting rainfall types using indigenous ecological indicators, built for the Zindi RAIL (Responsible Artificial Intelligence Lab) challenge. ## 📋 Challenge Overview **Goal:** Build a complete classification pipeline to predict the type of rainfall (Target) in the next 12–24 hours using indigenous ecological indicators and rainfall data. **Evaluation metric:** Macro F1 Score **Output:** submission.csv with columns [ID, Target] **Model format required:** ONNX (for eligibility) **Explainability requirement:** Include SHAP visualizations that highlight the most influential features for predictions. ## 🚀 Quick Start ### 1. Install Dependencies ```bash pip install -r requirements.txt ``` ### 2. Run the Complete Pipeline Open and run the Jupyter notebook: ```bash jupyter notebook indigenous_weather_forecasting_complete.ipynb ``` Or run it directly: ```bash jupyter nbconvert --to notebook --execute indigenous_weather_forecasting_complete.ipynb ``` ## 📊 Pipeline Features ### ✅ Complete Implementation 1. **Data Loading & EDA** - Comprehensive exploratory data analysis 2. **Feature Engineering** - Categorical encoding and numeric scaling 3. **Stratified Train-Validation Split** - Maintains class balance 4. **XGBoost Model Training** - Strong baseline classifier 5. **Model Evaluation** - Macro F1 score and detailed metrics 6. **SHAP Explainability** - Multiple visualization types 7. **Test Predictions** - Submission file generation 8. **ONNX Export** - Model serialization for production 9. **ONNX Validation** - Prediction consistency testing 10. **Feature Analysis** - Top 5 most important features with interpretations ### 📈 Generated Outputs The pipeline generates the following files: #### 📄 Submission Files - `submission.csv` - Final predictions for test set #### 🧠 Model Files - `model.onnx` - Trained model in ONNX …