Comparison of SVM and BPNN Estimation Models for Satisfaction in Old-Age Security

Abstract

Carrying out investigations into the relationships between the satisfaction in old-age security and its influence factors is of great significance for safeguarding social fairness and justice. As powerful statistical tools in machine learning, the support vector machine (SVM) and back-propagation neural network (BPNN) algorithms are used to develop nonlinear estimation models for satisfaction in the old-age security in China. Five influence factors (educational background, educational satisfaction, satisfaction with family’s financial situation, overall life satisfaction, society overall evaluation) were used as the input features. A SVM model obtained in this paper has prediction accuracies of 78.0% for the training set and 77.5% for the test, and a BPNN model possesses prediction accuracies of 77.8% and 77.0% for the two tests. Obviously, the SVM is superior to the BPNN in predicting satisfaction of old-age security in China.