Predictive Analysis for Personal Loans by Using Machine Learning

This study adopts five common machine learning algorithms for predicting consumer personal loan uptake, including Logistic Regression, Support Vector Machine, Multilayer Perceptron, Gradient Boosting Decision Trees Catboost, and Xgboost. The research utilizes data from Thera Bank available in the public database Kaggle, featuring fields like age, work experience, income, family size, average credit card expenditure, education level, home loans, securities account, deposit account, and internet banking usage. The study addresses the issue of imbalanced data using the SMOTE (Synthetic Minority Over-sampling Technique) method and compares the accuracy and stability of predictions using the five models with three different sampling rates to identify the optimal model and key factors. Empirical results show that the Gradient Boosting Catboost model and the Support Vector Machine model perform with stability and precision across different sampling ratios, making them the best models. Moreover, through the Gradient Boosting Xgboost model, the study identifies key features such as educational factors, income, family size, the existence of a deposit account, and annual credit card spending. The findings of this research can provide crucial factors for financial institutions when formulating marketing strategies for personal loans.