Predicting Personal Loan Acceptance: A Propensity Modelling Approach Using Logistic Regression and Machine Learning

Abstract With the cost of assets increasing steadily, purchasing high-value assets through personal savings has become difficult for many individuals. As a result, personal loans have emerged as a common and practical financial solution. For banks, identifying customers who are most likely to accept personal loans has become an important challenge. From a social-science and management perspective, understanding customer borrowing behaviour can help financial institutions make better, more inclusive decisions. This study focuses on developing and comparing different predictive models to identify customers who are likely to accept a personal loan. The data used for this research was obtained from Kaggle and consists of information on 5,000 customers of Thera Bank, with 14 demographic and financial attributes. The dataset contains no missing values and includes both numerical and encoded categorical variables. Logistic Regression, K-Nearest Neighbours (KNN), and Gaussian Naïve Bayes models were applied to predict personal loan acceptance. Model performance was evaluated using accuracy, precision, sensitivity, and ROC curves. The results show that the KNN model performs better than the other models, achieving the highest accuracy of 96.79%, followed by Logistic Regression and Gaussian Naïve Bayes. The KNN model also records higher sensitivity and precision, making it more effective in identifying customers who are likely to accept a personal loan. Overall, this study demonstrates how modern analytical techniques can support banks in targeting potential customers more effectively, increasing loan conversion rates, and improving decision-making in the financial sector (Han et al., 2012; Aggarwal, 2015).