Systematic and Flexible Genetic-Algorithm-Based Feature Reduction for Decision Tree ML-Validation

Abstract

In this paper, we propose a systematic genetic-algorithm-based feature reduction method. It has a high design flexibility based on 5-tuple parameter adjustment. The users can decide these 5 parameters to satisfy the demands of making the focus on accuracy or reduced feature amount. The proposed algorithm is verified by decision-tree models with different data sets. As for the data set, ala, the number of features is reduced from 123 to 53 while the accuracy performance has an increase of 4.2%. In addition, for other data sets, the maximum accuracy loss is no more than 3.1% while the feature reduction ratio achieves 41.9%. Its advantage is to provide a design trade-off between accuracy and reduced feature amount.