Dimensionality Reduction Using Hybrid Brainstorm Optimization Algorithm

In this work, a swarm intelligence-based algorithm, brainstorm optimization, is proposed for reducing dimensionality (feature selection) in datasets that are used for classification. Dimensionality reduction is a well-known and widely used technique in analyzing big data. Its role is to reduce the number of features in high-dimensional datasets and to keep only those that contain useful and rich information. This results in better understanding and interpretation of data, higher accuracy, and boosting the training process of machine learning method used for classification. After extracting features from the dataset, it should be decided which subset of features will be used in the training process. Since, in high-dimensional datasets many features exist, this problem is categorized as NP hard and it is necessary to employ metaheuristics for its solving. For tackling this issue, a binary hybrid brainstorm optimization metaheuristics that overcome the drawbacks of original algorithm, is proposed. For performance evaluation, 21 datasets are used. The comparative analysis is made between the proposed approach and the original brainstorm optimization algorithm, as well as with nine other metaheuristics adopted for feature selection. Experimental results prove the robustness of proposed method, since it is capable to reduce the number of features by simultaneously achieving better classification accuracy than other methods taken for comparative analysis.