Hybrid Feature Selection Method with Multi-objective Grey Wolf Optimizer for High Dimension Data

Abstract

Feature selection is an indispensable activity in machine learning, which aims at identifying the relevant predictors from a very high dimensional feature space to improve the performance and reduce the learning time of the model. However, a stupendous surge in the feature dimension space possesses a significant challenge to feature selection methods. This study is performed with an aim to cater to this challenge, and we propose a hybrid feature selection method composed of a metaheuristic optimization method and a filter-based feature selection. The proposed strategy cherry picks the advantages of both strategies while minimizing the overall drawbacks. We have examined and evaluated the proposed model over three benchmark datasets provided by University of California, Irvine Open-Source Repository. We benchmarked the performance and the accuracy of the proposed work against the standard feature selection methods and the existing hybrid models proposed in preceding studies. Adaptive boosting is used to assess the classification accuracy of the model. The set of standard classification metrics, such as, Accuracy, F1-Score, Recall and Precision are evaluated across all the three datasets for different feature selection strategies. The study result unanimously concludes that the proposed model outclasses the present standard and other state-of-art feature selection methods. The proposed model delivers accuracy of 96.60% over 95.68% for Ionosphere dataset. For the Sonar dataset, our model gives accuracy of 85.71% over 81.11%. For the Clean dataset, the proposed model's precision and recall are 96.31% and 95.10% over 95.01% and 94.93% respectively.