マ八ラノビス・タグチ・システムにおける実際の誤判別率

概要

論文の詳細を見る
The Mahalanobis-Taguchi System (MTS) has been proposed as a forecasting method using multivariate data for asymmetric discrimination problems such as the coin classification in a vending machine. In the situation, the normal coins constitute a separate population but the false coins do not. In this article, not only the unconditional but also the conditional error rates of a future observation from a normal group are investigated. If a nominal error rate is set by using chi-square distribution, the expected error rate will be much larger than the nominal value and the conditional error rates fluctuate around the expected value in the small sample cases. On the other hand, if the nominal error rate is set by using F distribution, the expected error rate coincides with the nominal value and the conditional error rates tend to be stable. Since it is difficult to estimate the conditional error rates based on an initial sample, it is advisable to set nominal values based on F distribution.
社団法人日本品質管理学会の論文
2007-01-15