A metaheuristic optimisation algorithm-based optimal feature subset strategy that enhances the machine learning algorithm’s classifier performance

Abstract

In machine learning feature selection is a powerful stage of choosing a subset of features that are useful to increase performance while decreasing dimensionality. The rule of thumb in selecting feature subsets in classifiers is proposed in this paper using a new metaheuristic optimization algorithm, which intends to enhance classifier performance. The proposed method takes advantage of metaheuristic algorithms to better search and select the most important features that contribute to increasing classification performance, decreasing overfitting and increasing of speed of computation. We coordinate the optimization process with the diverse machine learning classifiers such as SVM, Random Forests, and Neural Networks to compare the performance of the chosen feature subsets. The current gist of the paper shows that benchmark results on suitable datasets show the outperformance of the proposed strategy over regular feature selection procedures, hence leading to enhanced classifier performance. Therefore, this research forms part of the existing knowledge in feature selection for improving classification performances in various machine learning algorithms by offering a reliable approach for determining and applying the best relevant features.