Implementasi Algoritma Naive Bayes Untuk Prediksi Kelayakan Kredit Berbasis Status Pembayaran Pada BMT

Abstract

Credit assessment at microfinance institutions requires a consistent mechanism to identify applicants with a higher risk of payment problems. This study implements the Naïve Bayes classification algorithm in a credit prediction application at BMT . The study uses a case-study approach supported by observation and literature study, while the system is developed using the Waterfall model. Seven categorical predictors are used: gender, age group, occupation, loan amount group, repayment period, collateral, and income group, with payment status as the target class. The documented training set contains 10 records consisting of six “Lancar” and four “Macet” records. A reproducibility audit using categorical Naïve Bayes with Laplace smoothing was conducted because the manual probability example in the original report was not fully reproducible from the displayed training table. For the documented test pattern, the normalized posterior probabilities are 43.43% for Lancar and 56.57% for Macet, resulting in a Macet prediction. An additional leave-one-out cross-validation analysis on the 10 records produced 70.00% accuracy, with 66.67% precision, 50.00% recall, and 57.14% F1-score for the Macet class. User Acceptance Testing also passed all six tested scenarios. The findings indicate that Naïve Bayes can function as a decision-support component, but the small dataset limits generalization and requires further validation using a larger historical dataset.