Fuzzy Support Vector Machines based on Adaptive Particle Swarm Optimization for Credit Risk Analysis

Abstract

Every bank has loaning activities. Banks have several criteria for determining whether credit will give or not because every credit loan has a risk that the credit cannot return. In other words, banks need to analyze the credit applicant before granting the loan. Credit loan is a case of binary classification. The classification from applicant's data might be helpful for the bank in consideration whether the applicant will return the loan or not. Support Vector Machines (SVM) is a classification technique based on structural risk minimization, which is effective for binary classification. This method developed into Fuzzy Support Vector Machines (FSVM), which is able to minimize the influence of outlier in finding the best hyper plane. Adaptive Particle Swarm Optimization (APSO) is an extension of Particle Swarm Optimization (PSO). In APSO-based FSVM, APSO used to determine the fuzzy score by finding the class canter of each attribute that may give the highest accuracy. The result of this study is APSO-based FSVM can give the highest accuracy for each process. The highest rate of accuracy is 75,67%, which used APSO-based FSVM with 70% of training data and linear kernel.