Health Insurance Claim Classification using Support Vector Machine with Velocity Pausing Particle Swarm Optimization

Abstract

classification is a serious problem. Identifying claim classification is difficult. Machine Learning (ML) can predict potential claim decisions. Support Vector Machine (SVM) is a ML model that can generalize well to test data. SVM achieves an -score of 73.39% and 89.88% with a linear kernel, 73.34% and 73.34% with Radial Basis Function (RBF) kernel. Particle Swarm Optimization (PSO) improves the performance, because it can find the best parameters for SVM. However, the SVM parameters found by PSO are not guaranteed to be the global optimum. Velocity Pausing PSO (VPPSO) can address this problem. SVM-VPPSO performs better compared with SVM and SVM-PSO. SVM-VPPSO with linear kernel achieves -score of 90.17%, 90.16%, and 90.06% with 10, 20, and 30 particles respectively. The linear kernel also performs better than RBF kernel with a difference of 0.39% on the testing data. The best configuration is SVM-Linear-VPPSO with 10 particles. This configuration also achieves computation time of 46.938 seconds, which faster than SVM-Linear-VPPSO with 20 particles. The variance in computational time with 10 particles is 1.832 seconds, which better than with 20 particles with variance of 37.909 seconds.