PENYEIMBANGAN DATA PADA KLASIFIKASI DENGAN SUPPORT VECTOR MACHINE TERHADAP DATA PEMBAYARAN PINJAMAN BANK

Loan default in banking can cause losses. Therefore, lenders need to predict the criteria of customers who fail to pay their loans. In this study, a classification model was built to predict customers who fail to pay bank loans by applying the Support Vector Machine algorithm, especially with the Radial Basis Function (RBF) kernel. Because the occurrence of default is not balanced with the occurrence of smooth payments, a data balancing process was carried out. This study also compared the effect of data balancing methods using Random Over Sampling and Near Miss techniques on the performance of the SVM algorithm. The dataset used is the Loan Default Prediction Dataset taken from the kaggle site, which consists of 255,347 records and 18 attributes. The results showed that the SVM model trained without data balancing had the highest accuracy of 88.49%, but with a recall of only 10% and an F1-score of 17%. After using ROS, the model accuracy decreased slightly to 83.52%, but the recall increased significantly to 94% and the F1-score to 89%. With Near Miss, the model accuracy drops further by 65.29%, but produces better precision and recall compared to without balancing. It can be concluded that balancing with ROS provides the best performance in terms of the balance between precision and recall, as seen from the highest F1-score value among the three methods.