마할라노비스 다구찌 시스템의 다중 클래스 분류 성능 향상을 위한 새로운 방법

A Method for Improving Multiclass Classification Performance of Mahalanobis Taguchi System

Classification between normal and abnormal status of system is one of the important problems in all the industry. Mahalanobis Taguchi System(MTS) that is one of the pattern classification techniques based on mahalanobis distance concept and taguchi quality engineering theory is very efficient for singleclass classification problems; however, it has limitation of accuracy in multiclass classification problems. To overcome this limitation, this study proposes Simultaneous MTS(S-MTS) expanding the concept of the MTS. In the MTS, a single mahalanobis space of normal group is established; however, several mahalanobis spaces per reference group are established in the S-MTS. To validate the performance of the S-MTS, we applied the MTS and the S-MTS to the iris classification problem that is one of the most commonly used pattern classification problems. As a result, the MTS and S-MTS show 96% and 100% of classification accuracy respectively.