Development of evolved regression discriminant analysis combined ensemble learning for ordered categorical data

Abstract

In the fields of life sciences such as drug development, halting development of drug increases because of onset of toxicity recently. Therefore, rough toxicity prediction by using quantitative structure-activity relationship for ordered categorical data becomes more important. However we find difficulty in accurately analyzing ordered categorical data with existing methods because they treat their objective parameters as nominal instead of ordered. There are two analyses for ordered categorization: logistic regression analysis (LRA) and regression discriminant analysis (RDA). The RDA method has the following advantages over the liner discriminant analysis (LDA). 1) It provides important information in selecting explanatory variables. 2) It returns results more quickly. The RDA method, however, is not frequently utilized because of two weak points. In previous study, we refined the RDA method and developed the Evolved RDA method (ERDA). In this study, we tried to develop model supporting accuracy and generalization capability at the same time by using the ERDA method combined ensemble learning for ordered categorical data.