Evaluating Models Performance for Credit Risk Detection for Imbalanced Data

This study looks at the effect of unbalanced datasets on machine learning models and assesses the effectiveness of the Synthetic Minority Over-sampling Technique (SMOTE) in resolving this issue. The study demonstrates early variances in performance indicators using Naive Bayes, Decision Tree, and Logistic Regression models. SMOTE results in significant gains, notably in accuracy for Naive Bayes and Decision Tree, and precision for Logistic Regression. The findings provide nuanced impact of customized oversampling and the balancing strategies to overcome the drawbacks and threats of imbalanced datasets, leading to optimal model performance.