KOMPARASI ALGORITMA C4.5 (DECISION TREE) DAN NAIVE BAYES BERBASIS PARTICLE SWARM OPTIMIZATION UNTUK KLASIFIKASI KELAYAKAN KREDIT DEBITUR

Abstract

Credit financing across microfinance institutions and credit cooperatives drives regional economic mobility but remains vulnerable to non-performing loan defaults. Manual appraisal mechanisms frequently suffer from subjective biases and error-prone assessments. This study comparatively examines C4.5 (Decision Tree) and Gaussian Naïve Bayes classification algorithms, enhanced through feature selection driven by Particle Swarm Optimization (PSO). Utilizing an empirical dataset of 1,200 loan applications encompassing 14 financial and demographic attributes, predictive accuracy was validated via 10-fold cross-validation across Accuracy, Precision, Recall, F-Measure, and ROC-AUC parameters. Experimental benchmarks show that PSO feature selection pruned 5 redundant features, boosting predictive metrics across both paradigms. The PSO-optimized C4.5 framework attained the peak accuracy of 91.42% and an AUC of 0.948, significantly outperforming baseline C4.5 (86.25% accuracy, 0.891 AUC) and PSO-Naïve Bayes (88.17% accuracy, 0.915 AUC). This demonstrates that swarm intelligence optimization combined with decision trees furnishes a dependable decision-support model to mitigate debtor credit insolvency.