Aa0414/Kenya-Loan-Default-Prediction

This project focuses on building a machine-learning model to predict whether a borrower will default on a credit-card loan. Using the Zindi credit-risk dataset, we explored the data, engineered features, and trained multiple classification models to understand which approaches performed best. # Kenya-Loan-Default-Prediction Team 6: Bruce Liu, Grace Kung, Emily Su, Allen Chiu BA810 Supervised Machine Learning (Fall 2025) Boston University, Questrom School of Business This project focuses on building a machine-learning model to predict whether a borrower will default on a credit-card loan. Using the Zindi credit-risk dataset, we explored the data, engineered features, and trained multiple classification models to understand which approaches performed best.Our evaluation combined standard accuracy metrics with a cost-sensitive framework to reflect the real financial consequences of misclassifying borrowers. Because the cost matrix was based on estimated values, the results highlight both the potential of cost-aware modeling and the need for real-world cost data in future iterations. The model performs reasonably well in distinguishing high-risk and low-risk borrowers, offering useful guidance for lenders seeking to reduce losses while maintaining fair access to credit. At the same time, the analysis reveals challenges related to data quality, feature interactions, and model generalization, all of which influence overall performance. These findings emphasize the importance of careful data handling, ongoing model tuning, and responsible evaluation when applying machine learning to financial decision-making. Overall, this project demonstrates how predictive modeling can support lenders, borrowers, and government stakeholders in creating a more stable, fair, and informed credit system. The results serve as a strong foundation for further improvements, including the use of real operational cost data, more advanced feature engineering, and broader evaluation strategies.