ANALISIS KOMPARASI ALGORITMA RANDOM FOREST, XGBOOST, DAN MULTILAYER PERCEPTRON (MLP) DALAM PREDIKSI RISIKO GAGAL BAYAR KREDIT

Abstract

Credit risk prediction is a critical task for financial institutions in identifying customers at risk of default. This study compares the performance of three machine learning and deep learning algorithms as in Multilayer Perceptron (MLP), Random Forest, and XGBoost in predicting credit card defaults using the “Default of Credit Card Clients” dataset from the UCI Machine Learning Repository. The dataset consists of 30,000 records with 23 features covering demographic information, payment history, bills, and payment amounts over a six-month period. Class imbalance was addressed using the Synthetic Minority Oversampling Technique (SMOTE), which was applied only to the training data. Model performance was evaluated using accuracy, precision, recall, F1-score, and AUC-ROC metrics. The experimental results show that Random Forest achieved the best overall performance with an F1-score of 0.5275 and an AUC-ROC of 0.768, outperforming MLP (F1-score 0.5182, AUC-ROC 0.7610) and XGBoost (F1-score 0.5080, AUC-ROC 0.7616%). These findings indicate that ensemble-based methods remain competitive compared to deep learning approaches for tabular credit data, and provide valuable insights for financial institutions in implementing data-driven risk management.