Analysis and Comparison of Classification Algorithms for Credit Approval in Islamic Banks

Abstract

In the context of Islamic banking, an efficient and quick credit analysis process is crucial to meet customer needs without compromising Sharia principles. Although Islamic banks do not charge interest, profits are obtained through contracts, requiring thorough and accurate credit analysis. This study evaluates the performance of various classification algorithms, including C4.5, Random Forest, Decision Tree, K-Nearest Neighbors, and Naïve Bayes, in predicting credit approval decisions based on factors such as character, capacity, capital, collateral, and economic conditions. Using a dataset from an Islamic bank, the algorithms were assessed through precision, sensitivity, and accuracy metrics. The results highlight significant performance differences among the algorithms. Random Forest and Decision Tree demonstrated strong training accuracy but suffered from overfitting, limiting their generalization to new data. Conversely, the C4.5 algorithm achieved a balanced performance with a testing accuracy of 74.5%, making it a promising candidate for practical application in Sharia-compliant credit risk assessment. This study emphasizes the importance of selecting algorithms that address overfitting while maintaining robust predictive accuracy, contributing to improved decision-making in Islamic banking.