Feasibility Classification of New Design Points Using Support Vector Machine Trained by Reduced Dataset

Abstract

In this paper, we propose to use a support vector machine (SVM) for the classification of design data. Although the SVM is a very popular technique in data mining, it is rarely applied to an industrial design process that may require information regarding the feasibility of the design point of interest. To check the feasibility, the designer must conduct experiments or computer simulations, which may incur considerable cost. Therefore, the SVM can be an effective tool for predicting feasible and infeasible regions because it only uses the cumulative design data. In this paper, we used the SVM to classify sample datasets drawn from mathematical test problems and from an air-conditioner pipe design example. Our results indicate that the SVM is capable of very accurately identifying feasible and infeasible regions in the design space. Further, we were able to reduce the training time of the SVM by using the k-means clustering algorithm to reduce the amount of training data, taking advantage of the powerful generalization abilities of the SVM. Consequently, we conclude that the SVM can be an effective tool to assess feasibility at certain design points, avoiding some of the high computational costs of the analysis.