Demographic and Financial Trends in the Housing Market: A Machine Learning Approach to Mortgage Loans in Federal Home Loan Banking

This poster presents an analysis of mortgage loan data from the Federal Home Loan Bank (FHLB) system to identify trends related to borrower demographics, loan characteristics, and lending practices. After exploring various financial and demographic patterns, we selected borrower credit score as the target variable due to its central role in lending decisions and its strong correlation with multiple financial indicators. To model and classify credit score tiers, we implemented a machine learning pipeline focused on both predictive performance and interpretability. A two-stage feature selection process—Mutual Information followed by Recursive Feature Elimination with Cross-Validation (RFECV)—was used to identify the most relevant predictors. Eight core features, including income, debt ratios, borrower age, and socioeconomic indicators, were used to train tree-based classifiers (Random Forest, Decision Tree, Extra Trees). Models were evaluated using an 80/20 train-test split and 10-fold cross-validation. The best-performing model, Extra Trees, achieved an accuracy of 79.73% under 10-fold CV. Most misclassifications occurred in the mid-range credit score categories, while the highest tier (750+) showed minimal error—likely reflecting a bias toward high-credit borrowers in lending practices within the FHLB system. Geographic trends in lending were also explored to inform future spatial analyses. Presented at the Kansas Data Science Conference 2025.