TY - JOUR AU - Shuichi Shinmura PY - 2016/06/01 Y2 - 2024/03/29 TI - The Best Model of the Swiss Banknote Data -Validation by the 95% CI of coefficients and t-test of discriminant scores JF - Statistics, Optimization & Information Computing JA - Stat., optim. inf. comput. VL - 4 IS - 2 SE - Research Articles DO - 10.19139/soic.v4i2.178 UR - http://iapress.org/index.php/soic/article/view/20160603 AB - The discriminant analysis is not the inferential statistics since there are no equations for standard error (SE) of error rate and discriminant coefficient based on the normal distribution. In this paper, we proposed the “k-fold cross validation for small sample” and can obtain the 95% confidence interval (CI) of error rates and discriminant coefficients. This method is the computer-intensive approach by statistical and mathematical programming (MP) software such as JMP and LINGO. By the proposed approach, we can choose the best model with the minimum mean of error rate in the validation samples (Minimum M2 Standard). In this research, we examine the sixteen linear separable models of Swiss banknote data by eight linear discriminant functions (LDFs). M2 of the best model of Revised IP-OLDF is the smallest value of all models. We find all coefficients of six Revised IP-OLDF among sixteen models rejected by the 95% CI of discriminant coefficients (Discriminant coefficient standard). We compare t-values of the discriminant scores. The t-value of the best model has the maximum values among sixteen models (Maximum t-value Standard). Moreover, we can conclude that all standards support the best model of Revised IP-OLDF. ER -