Enhancing Credit Scoring Models with Monte Carlo Simulated Features

Credit scoring is an essential tool in financial risk management, influencing lending decisions and the broader economic landscape. Traditional credit scoring methods often fail to capture complex behavioral patterns. This paper explores the enhancement of credit scoring models through the integration of Monte Carlo simulated features, aiming to reveal latent patterns within credit data. We implement a comparative analysis across four distinct scenarios: utilizing original features, new synthetic features, a combination of both, and a refined model excluding highly correlated features. Our methodology employs a robust feature engineering process using Monte Carlo simulations to generate synthetic data, enriching the feature space. We evaluate the performance of conventional machine learning models-Logistic Regression, Decision Tree, Random Forest, and Support Vector Machine-in each scenario, adopting accuracy, precision, recall, F1 score, and ROC AUC as our metrics. The results demonstrate a significant improvement in predictive accuracy when synthetic features are included, suggesting that such features can capture nuanced risk factors in credit behavior. Our findings show the potential of synthetic feature integration in developing more insightful and reliable credit scoring models.