kpatc/financial_health_index_prediction

Zindi competition to predict financial health (Low/Medium/High) for small and medium-sized businesses in Southern Africa. --- # Financial Health Index Prediction A machine learning pipeline to predict financial health (Low/Medium/High) for small and medium-sized businesses in Southern Africa. This Zindi competition project uses gradient boosting to classify 9,600+ SMEs based on their business metrics, financial behaviors, and operational characteristics. **Dataset**: 9,618 training records × 39 features | 2,405 test records **Target**: 3-class classification (Low: 65%, Medium: 30%, High: 5% — severe imbalance) **Best Model**: Gradient Boosting (F1: 0.82, Accuracy: 78.7%) **Status**: ✅ Production-ready with API deployment --- ## Executive Summary **Key Finding**: Financial metrics drive health classification, NOT demographics. - **Top Predictor**: Business Turnover (18% importance) - **Second**: Personal Income (15%) - **Third**: Business Expenses (12%) - **Critical Insight**: Financial metrics account for 75% of predictive power. Business size and profitability dominate health assessment. --- ## Quick Start ### 1. Setup Environment ```bash source venv/bin/activate pip install -r requirements.txt ``` ### 2. Run Full Pipeline ```bash # Phase 1: Exploratory Data Analysis jupyter notebook eda/eda.ipynb # Phase 2: Data Cleaning python cleaning/clean.py # Phase 3: Model Training python modeling/train.py # Phase 4: Generate Test Predictions python modeling/predict.py # Output: test_predictions.csv ``` --- ## Project Structure ``` Financial_health_index/ ├── README.md # This file ├── requirements.txt # Python dependencies │ ├── data/ │ ├── Train.csv # 9,618 records × 39 features + Target │ ├── Test.csv # 2,405 records × 38 features │ ├── VariableDefinitions.csv # Feature metadata & descriptions │ └── cleaned/ # Cleaned data outputs (generated) │ ├── eda/ │ └── eda.ipynb # 8-section exploratory analysis │ ├── cleaning/ │ ├── __init__.py │ └── clean.py …