Prediksi Nasabah Gagal Bayar Pinjaman (Default) Menggunakan Algoritma Random Forest

Abstract

Getting a loan from a financial institution has become a common sight in today's life. Every day, there are many of people apply for loans for various purposes. But not all loan applicants are reliable and not everyone gets approved. Every year there are cases where some people are found to have failed to repay their loans and this results in huge financial losses for the loan provider. Therefore, this research aims to predict loan defaults. The method used is random forest which has an ensemble learning type, which is a technique of combining several models to predict default customers or non-default customers. This method has resistance to outliers, so that the accuracy in predicting is correct without being affected by the presence of outliers. So in this study the outliers were ignored. The performance of the random forest model in this research case study has good performance results, namely with an accuracy of 96% obtained based on the evaluation metric with confusion matrix and ROC curve. Then also obtained the best tuning hyperparameter value of random forest in this study, namely n_estimator = 80 and max_depth = 10. With the model that has been produced, it is hoped that lending institutions can speed up the process more efficiently and also with good performance to determine default and non-default customers.