Enhancement of Credit Score Prediction Using Artificial Neural Network with Adaptive Particle Swarm Optimization

Abstract

Credit scoring is a critical and most important process for every financial and lending institution to make informed lending decisions and effective risk management. Effective credit risk assessment plays a crucial role in minimizing default risk and enhancing recovery assurance, thereby strengthening the confidence of lending institutions in making informed and responsible loan decisions. Historically, traditional statistical methods such as logistic regression, linear discriminant analysis, and decision trees have been used to classify credit applicants. Still, these methods failed to handle non-linear data of the current economic landscape. Machine learning algorithms have improved credit scoring by handling non-linear data more effectively. However, they often struggle with tuning hyper-parameters, highlighting the need for more advanced and efficient models to manage complex data. This study aims to enhance credit score prediction using ANN with APSO. Three models namely: ANN without enhancement, ANN enhanced with the PSO, and ANN enhanced with the APSO were developed and evaluated using a popular real-world German credit dataset which was further split into training, validation, and testing datasets to assess the performance of the developed model with unseen dataset. Performances of the models were assessed using accuracy, precision, recall, F₁-score, and F_0.5-score.  Additionally, a confusion matrix, loss curve, and precision-recall (PR) curve were generated for detailed analysis. The ANN enhanced with the APSO model achieved an accuracy of 80.00%, precision of 84.11%, F1-score of 85.71%, and F0.5-score of 84.74%, demonstrating superior performance over both the baseline ANN and ANN enhanced with the PSO. Specifically, ANN enhanced with the APSO improved the F1-score by 5.7% over the baseline ANN, and by 3.4% over ANN enhanced with the PSO, highlighting its enhanced effectiveness. Compared to optimization methods like Grid Search (GS) and Genetic Algorithm (GA), ANN-APSO also showed more consistent and higher performance metrics. These results suggest that the proposed ANN enhanced with the APSO model offers a more accurate and robust approach to credit scoring, with the potential to significantly improve risk assessment and support more informed, lower-risk lending decisions for financial institutions.