zameer-Gimba/Loan-Default-Risk-Predictor

Final Capstone Project for the 3MTT (AI/ML track). This repository leverages gradient boosting algorithm (LightGBM) to model and predict credit default risks using the Data Science Nigeria (DSN) Challenge 1 digital lending dataset from Zindi. # Loan Default Risk Predictor A machine learning project that predicts loan default risk using the Data Science Nigeria (DSN) Challenge 1 dataset from Zindi. ## Dataset - `traindemographics.csv` - `trainperf.csv` - `trainprevloans.csv` ## Model - LightGBM Classifier - 5-fold Stratified Cross-Validation - Holdout test split: 80/20 ## Metrics - F1-Score - Recall - PR-AUC - ROC-AUC - Precision - Balanced Accuracy ## Project Structure - `src/` — data prep, feature engineering, training, prediction - `app/` — Streamlit app - `models/` — saved model artifact - `test.py` — unit tests ## Run Training ```bash python -m src.train ``` ## Run App ``` streamlit run app/streamlit_app.py ``` ## Run Tests ``` python test.py ``` ## Run Order ```bash python -m src.train streamlit run app/streamlit_app.py python test.py ``` Streamlit: loan-default-risk-predictor…)