drustagi1647/M-Shwari-project

Predicting M-Shwari mobile credit uptake in Kenya using XGBoost and SHAP — with fairness analysis across gender, wealth, and age. # M-Shwari-project Predicting M-Shwari mobile credit uptake in Kenya using XGBoost and SHAP — with fairness analysis across gender, wealth, and age. # M-Shwari Credit Scoring — Kenya FinAccess 2016 This project builds a machine learning pipeline to predict uptake of M-Shwari mobile banking credit among Kenyan households, using the FSD Kenya FinAccess 2016 national survey (8,665 respondents). The analysis is motivated by Bharadwaj, Jack & Suri (2019), which showed that M-Shwari loans improve household resilience to financial shocks, but treated the underlying credit-scoring algorithm as a black box. This project opens that black box — identifying which observable household and behavioral characteristics best predict who takes up this kind of mobile credit. ## Key Findings - XGBoost model achieves AUC = 0.95 on a held-out test set - Mobile banking usage is the single strongest predictor by a wide margin (SHAP = 6.57), followed by formal savings status and household wealth index - Fairness analysis reveals the model performs better for women than men, best for the poorest wealth quintile, and fails entirely for respondents over 60 due to data sparsity ## Pipeline - Data loading from SPSS (.sav) format via pyreadstat - Feature engineering: 20 features across demographics, wealth, digital access, credit behavior, and savings behavior - Class imbalance handled with SMOTE (4.9% positive rate) - XGBoost classifier with threshold tuning (optimized at 0.10 for recall) - SHAP TreeExplainer for interpretability - Subgroup fairness analysis across gender, wealth quintile, and age group ## Dataset FSD Kenya FinAccess 2016 — available at fsdkenya.org ## Reference Bharadwaj, P., Jack, W., & Suri, T. (2019). Fintech and Household Resilience to Shocks: Evidence from Digital Loans in Kenya. NBER Working Paper No. 25604.