A Feature Selection Approach Based on Improved Binary Coyote Optimization Algorithm

Feature selection is one of the essential approaches for machine learning and data mining to eliminate the irrelevant and redundant features from the datasets. As the wrapper-based feature selection can be considered as a binary optimization problem, an improved binary coyote optimization algorithm (IBCOA) is proposed and applied for solving the feature selection problems in this paper. In the proposed algorithm, two Sigmoidal transfer functions are utilized respectively for updating the social condition of each coyote and dealing with the birth of new coyotes. An alpha based mutation strategy is applied for updating and binarizing the coyotes in a pack. The proposed approach is evaluated using twelve well-known datasets from the UCI machine learning repository. The experimental results confirm that the proposed approach outperforms the other two feature selection approaches based on the conventional COA and a V-shaped binary COA.