Mutually-exclusive-and-collectively-exhaustive feature selection scheme

In the fields of machine learning and data mining, feature selection methods are used to identify the most cost-effective predictors and to give a deeper understanding of pattern recognition and extraction. This study proposes a novel mutually-exclusive-and-collectively-exhaustive (MECE) feature selection scheme. Based on the MECE principle in decision science, the scheme, which has three stages including evaluation of independence, evaluation of importance and evaluation of completeness, aims to identify the independent and important variables with complete information. A case study of fault classification in semiconductor manufacturing and a study of breast cancer relapse identification in bioinformatics are used to validate the proposed scheme. The results demonstrate that the proposed MECE scheme selects fewer variables, avoids the multicollinearity problem, and improves fault classification accuracy in the two case studies. (C) 2017 Elsevier B.V. All rights reserved.