Analysis of metaheuristics feature selection algorithm for classification

Abstract

Classification is a very vital task that is performed in machine learning. A technique used for classification is trained on various instances to foresee the class labels of hidden instances, and this is known as testing instances. The technique used for classification is able to find the connection between the class and instances due to the aid from the training process known as attributes. Redundant and non-relevant data are eradicated from the dataset with feature selection technique and these gives room for enhancement of the classification performance through feature selection. This research displays the feature selection techniques performances and are divided into wrapped-based metaheuristics algorithm and filter-based algorithms using two educational datasets. Four different classification techniques were used on the datasets and the outcome shows that Decision Tree (DT) gave the best performance on the datasets. Furthermore, the result shows that the proposed CHPSO-DE outshined other feature selection algorithms in that it obtained the best classification performance by using fewer features. The result of the various feature selection and classification technique will help researchers in getting the most efficient of feature selection algorithms and classification techniques.