MOBCSA: Multi-Objective Binary Cuckoo Search Algorithm for Features Selection in Bioinformatics

Abstract

In bioinformatics, medical diagnosis models might be significantly impacted by high-dimensional data generated by high-throughput technologies. This data includes redundant or irrelevant genes making it challenging to identify the relevant genes from such high-dimensional data. Therefore, an effective feature selection (FS) technique can significantly reduce the degree of dimensionality to enhance the performance and accuracy of medical diagnosis. Cuckoo Search Algorithm (CSA) is applied for gene selection and found to be effective in terms of exploitation, exploration, and convergence. However, most of the current CSA-based FS techniques deal with gene selection problem as a single objective rather than a multi-objective mechanism. This article proposes a Multi-Objective Binary Cuckoo Search Algorithm (MOBCSA) for gene selection. The MOBCSA extends the standard CSA considering multiple objectives such as accuracy of classification and number of selected genes. MOBCSA utilizes S-shaped transfer function for transforming the algorithm’s search space from a continuous to a binary search space. MOBCSA integrates two components: an external archive to save the pareto optimal solutions attained during the search process and an adaptive crowding distance updating mechanism integrated into the archive to maintain diversity and increase the coverage of optimal solutions. To evaluate the performance of MOBCSA, the evaluation experiments were conducted on six benchmark biomedical datasets using three different classifiers. Then, the obtained experimental results were compared against four multi-objective-based state of the art FS methods. The findings prove that MOBCSA surpasses the other methods in both accuracy of classification and number of selected genes, where it has obtained an average accuracy ranging from 92.79% to 98.42% and an average number of selected genes ranging from 15.67 to 27.88 for different classifiers and datasets.