Conditional Mutual Information-based Feature Selection for Sex Differences Characterization

Abstract

This study proposes a feature selection approach exploiting Conditional Mutual Information to identify the most relevant features to perform sex classification. The approach applied to features extracted from cardiovascular time series, is combined with a Linear Discriminant Analysis classifier. The feature selection method allowed to noticeably reduce the number of used features, achieving at the same time comparable and acceptable accuracy (∼ 62%) and overall good recall and F1-scores for females (∼ 71% and ∼ 63%, respectively).