Klasifikasi Nasabah Kartu Kredit Menggunakan Algoritme Fuzzy K-Nearest Neighbor pada Data Tidak Seimbang

Abstract

Debtor classification is a crucial banking process in order to identify potential problematic credit card applicant. This information is to support decision making in accepting or rejecting the application. The classification method makes use of fuzzy k-nearest neighbor method. Due to the nature of credit card business, most likely, the data collected is imbalanced. In this case, the good debtors always significantly outnumber the bad one. Most of the existing classification systems work well on balanced data, which is not the case in this research. Thus, in order to improve the system, the data composition must be banced using oversampling and undersampling technique. The performance indicator used are accuracy, precision, recall, and f-measure. Replication oversampling improved the algorithm best at the number of nearest neighbors 1. The accuration, precision, recall, and f-measure is 91.93%, 86.12%, 100%, and 92.54%, while the least performance was achieved at the number of nearest neighbors 5. Random oversampling performed better as the number of the nearest neighbors increases. Undersampling performed more stable with the number of the nearest neighbors between 1 and 5.