An Evaluation of Mahalanobis-Taguchi System Experiment Metrics

Abstract

The Mahalanobis-Taguchi System (MTS) is a dimension reduction and prediction method for multivariate data. the method can be described by four main steps: construction of a full model scale, validation of the scale, experimentation to identify the useful variables, and application of the robust scale to diagnose, classify, or forecast. Dimension reduction is accomplished during the experiment stage of MTS where a metric is used to identify the useful variables. Traditionally, the larger-the-better (LTB) signal-to-noise ratio is the metric used for classification applications of MTS, since the true levels of severity are not known. However, if a level of severity is assigned to each class in the study, such as a one for normal and two for abnormal, the dynamic signal-to-noise ratio could also be used. This method of assignment also allows the use of the correlation coefficient as a possible metric. the goal of this research is to examine new metrics that identify useful variables during the experiment stage within MTS for a classification application. Each metric's ability to discriminate between two classes will be measured.

Department(s)

Engineering Management and Systems Engineering

Keywords and Phrases

Classification; Data mining; Mahalanobis distance (MD); Mahalanobis-Taguchi System (MTS); Signal to noise ratio

Document Type

Article - Conference proceedings

Document Version

Citation

File Type

text

Language(s)

English

Rights

© 202 Scimagp Journal and Country Rank, All rights reserved.

Publication Date

01 Jan 2011

This document is currently not available here.

Share

 
COinS