Beyond Fisher's Linear Discriminant Analysis – New World of Discriminant Analysis
Shuichi Shinmura
Economics, Seikei Univ., Musashino, Tokyo, Japan

LDF based on MNM (minimum number of misclassifications) criterion using integer programming (Revised IP-OLDF) is proposed because most of real data doesn't satisfy Fisher's assumption. Revised IP-OLDF was evaluated with LDF and the logistic regression by the 100 fold cross-validation of four different data such as Iris data, Swiss Bank Data (MNM = 0), CPD Data, Student Data. Comparing 13,500 results of error rate, most of 135 mean error rates by Revised IP-OLDF are superior to LDF and the logistic regression.

Keywords: Comparison of 135 mean error rates; linear discriminant function based on MNM criterion; Fisher's linear discriminant function; Logistic regression model

Biography: I introduced statistical and Mathematical programming software such as SAS, SPSS, JMP and LINDO products into Japan. I wrote over 10 books about statistics and MN for researcher and education of universities. I have many experiances about discriminant analysis in Medical field etc. I develope Revised IP-OLDF and evaluate 12 years ago. I finished it last Oct.