An improved multi-objective marine predator algorithm for gene selection in classification of cancer microarray data
Qiyong Fu,Qi Li,Xiaobo Li
DOI: https://doi.org/10.1016/j.compbiomed.2023.107020
IF: 7.7
2023-05-11
Computers in Biology and Medicine
Abstract:Gene selection (GS) is an important branch of interest within the field of feature selection, which is widely used in cancer classification. It provides essential insights into the pathogenesis of cancer and enables a deeper understanding of cancer data. In cancer classification, GS is essentially a multi-objective optimization problem, which aims to simultaneously optimize the two objectives of classification accuracy and the size of the gene subset. The marine predator algorithm (MPA) has been successfully employed in practical applications, however, its random initialization can lead to blindness, which may adversely affect the convergence of the algorithm. Furthermore, the elite individuals in guiding evolution are randomly chosen from the Pareto solutions, which may degrade the good exploration performance of the population. To overcome these limitations, a multi-objective improved MPA with continuous mapping initialization and leader selection strategies is proposed. In this work, a new continuous mapping initialization with ReliefF overwhelms the defects with less information in late evolution. Moreover, an improved elite selection mechanism with Gaussian distribution guides the population to evolve towards a better Pareto front. Finally, an efficient mutation method is adopted to prevent evolutionary stagnation. To evaluate its effectiveness, the proposed algorithm was compared with 9 famous algorithms. The experimental results on 16 datasets demonstrate that the proposed algorithm can significantly reduce the data dimension and obtain the highest classification accuracy on most of high-dimension cancer microarray datasets.
engineering, biomedical,computer science, interdisciplinary applications,mathematical & computational biology,biology