Evaluasi Kinerja Algoritma Naive Bayes, Decision Tree, dan Random Forest untuk Prediksi Risiko Default Nasabah Kartu Kreditult Nasabah Kartu Kredit

Sari

Abstrak: Prediksi risiko gagal bayar (default) nasabah kartu kredit merupakan aspek penting dalam manajemen risiko kredit pada lembaga keuangan. Berbagai penelitian telah menerapkan algoritma klasifikasi untuk prediksi risiko kredit, namun sebagian besar penelitian hanya berfokus pada satu atau dua algoritma sehingga hasil perbandingan performa antar metode masih terbatas. Penelitian ini bertujuan untuk mengevaluasi dan membandingkan kinerja algoritma Naive Bayes, Decision Tree, dan Random Forest dalam memprediksi risiko default nasabah kartu kredit. Dataset yang digunakan merupakan data sekunder dari platform Kaggle yang terdiri dari 30.000 data nasabah. Tahapan penelitian meliputi preprocessing data, pembagian training dan testing dengan rasio 80:20, pembangunan model klasifikasi, serta evaluasi menggunakan metrik accuracy, precision, recall, dan F1-score. Kontribusi penelitian ini terletak pada evaluasi komparatif tiga algoritma klasifikasi dalam satu kerangka eksperimen yang sama menggunakan pendekatan evaluasi multi-metrik. Hasil penelitian menunjukkan bahwa Random Forest menghasilkan kinerja terbaik berdasarkan nilai accuracy sebesar 0,813500 dan precision sebesar 0,630027. Sementara itu, Naive Bayes memperoleh nilai recall tertinggi sebesar 0,651181 dan F1-score sebesar 0,494363. Temuan penelitian menunjukkan bahwa Random Forest lebih unggul dalam ketepatan klasifikasi, sedangkan Naive Bayes lebih efektif dalam mendeteksi nasabah yang berpotensi mengalami gagal bayar.
Kata kunci: Credit Risk Assessment, Kartu Kredit, Prediksi Risiko Kredit, Naive Bayes, Decision Tree, Random Forest

Abstract: Credit card default risk prediction is an important aspect of credit risk management in financial institutions. Although various classification algorithms have been applied to credit risk prediction, most previous studies have focused on only one or two algorithms, resulting in limited comparative insights into their performance. This study aims to evaluate and com-pare the performance of Naive Bayes, Decision Tree, and Random Forest algorithms in pre-dicting credit card customer default risk. The dataset used in this study is secondary data ob-tained from Kaggle, consisting of 30,000 customer records. The research process includes data preprocessing, dataset splitting into training and testing sets with an 80:20 ratio, clas-sification model development, and performance evaluation using accuracy, precision, recall, and F1-score metrics. The contribution of this study lies in the comparative evaluation of three classification algorithms within the same experimental framework using a multi-metric evaluation approach. The results indicate that Random Forest achieved the best overall per-formance with an accuracy of 0.813500 and a precision of 0.630027. Meanwhile, Naive Bayes obtained the highest recall of 0.651181 and the highest F1-score of 0.494363. These findings suggest that Random Forest is more effective in overall classification performance, whereas Naive Bayes is better at identifying customers with a higher likelihood of default.
Keywords: Credit Risk Assessment, Credit Card, Credit Risk Prediction, Naive Bayes, Decision Tree, Random Forest