Credit Scoring

Credit scoring estimates the likelihood that people or firms will repay loans, guiding approvals, pricing, limits, and monitoring. Work ranges from logistic regression and boosted trees to neural and graph models, with newer approaches using transaction histories, alternative data, tabular foundation models, and federated learning. Researchers evaluate these systems on datasets such as LendingClub, Home Credit, mortgage records, and small-business portfolios, measuring default ranking alongside calibration, recall, robustness to shifts, interpretability, and group fairness. Predictive gains do not guarantee fair or useful decisions: proxy variables, selective labels, changing economic conditions, and explanation quality remain difficult, while privacy and governance constrain which data and adaptations can be used.