Abstract
Feature Selection (FS) has a non-trivial role in supervised learning; like classification, for many causes. FS aims at facilitating the model processes and reducing the computation time. In feature selection, trivial features are eliminated from the data to produce transparently and comprehensibly a model. Furthermore, a feature selection process can decrease noise data; wherefore, feature selection enhances the accuracy measure of the classification process. This paper proposes a robust hybrid dynamic model for feature selection, called RS-CBGWO-FS. RS-CBGWO-FS is a combination of Rough Set (RS), chaos theory and Binary Grey Wolf Optimisation (BGWO). GWO parameters are estimated and tuned by using ten various chaotic maps. Five complex medical data sets are used in the evaluation experiments. The selected data sets have various uncertainty attributes and missing values. The overall result indicates that RS-CBGWO-FS with the Singer and piecewise chaos maps provides better effectiveness, minimal error, higher convergence speed and lower computation time.