Journal of King Saud University: Computer and Information Sciences (Sep 2023)
A novel improved lemurs optimization algorithm for feature selection problems
Abstract
The irrelevant and repeated features in high-dimensional datasets can negatively affect the final performance and accuracy of classification-based models. Therefore, feature selection (FS) techniques can be used to determine the most optimal relevant features. In this paper, we fuse a new enhanced model from Lemurs Optimization (LO) algorithm, called Enhanced Lemurs Optimization (ELO). We combine Opposition Based Learning (OBL) and Local Search Algorithm (LSA) to address exploration and exploitation challenges, respectively. Our proposed ELO algorithm incorporates U-shaped and Sigmoid transfer functions during the position update step, leading to improved accuracy and convergence. These new deployments based on the U-shaped and Sigmoid transfer functions are called ELO-U and ELO-S algorithms, respectively. The performance of all three new versions of our proposed optimization algorithms (ELO, ELO-U, and ELO-S) has been evaluated using 21 UCI datasets in different fields and sizes. Moreover, their results are also compared to other competitive algorithms. The evaluation process included several measurements such as fitness value, an average of selected features, and average accuracy. Experimental results demonstrate that our proposed ELO-U algorithm achieves the best average accuracy of 91.03%. Statistical analysis using Friedman and Wilcoxon tests confirms the superiority of ELO-U over other competitors.
