Fig. 1.

Experimental design of this study. There are 17 datasets used to evaluate the performances of the feature selection algorithms, as in box of “Data Input”. The structure of proposed McTwo algorithm is illustrated in the box “McTwo”. Nine other representative feature selection algorithms are listed in the box “Other Methods”. Four binary classification algorithms are used to evaluate what degree of accuracy the classification models based on the selected features may achieve. The classification performance is measured by the sensitivity (Sn), specificity (Sp), overall accuracy (Acc). a The processing scheme of internal cross validation. b The work flow of external cross validation