A Binary Waterwheel Plant Optimization Algorithm for Feature Selection

Abstract

The vast majority of today’s data is collected and stored in enormous databases with a wide range of characteristics that have little to do with the overarching goal concept. Feature selection is the process of choosing the best features for a classification problem, which improves the classification’s accuracy. Feature selection is considered a multi-objective optimization problem with two objectives: boosting classification accuracy while decreasing the feature count. To efficiently handle the feature selection process, we propose in this paper a novel algorithm inspired by the behavior of waterwheel plants when hunting their prey and how they update their locations throughout exploration and exploitation processes. The proposed algorithm is referred to as the binary waterwheel plant algorithm (bWWPA). In this particular approach, the binary search space as well as the technique’s mapping from the continuous to the discrete spaces are both represented in a new model. Specifically, the fitness and cost functions that are factored into the algorithm’s evaluation are modeled mathematically. To assess the performance of the proposed algorithm, a set of extensive experiments were conducted and evaluated in terms of 30 benchmark datasets that include low, medium, and high dimensional features. In comparison to other recent binary optimization algorithms, the experimental findings demonstrate that the bWWPA performs better than the other competing algorithms. In addition, a statistical analysis is performed in terms of the one-way analysis-of-variance (ANOVA) and Wilcoxon signed-rank tests to examine the statistical differences between the proposed feature selection algorithm and compared algorithms. These experiments’ results confirmed the proposed algorithm’s superiority and effectiveness in handling the feature selection process.