BCOOT-Based Binary Optimization for Microarray-Based Cancer Gene Selection and Classification: A Comparative Study

Abstract

Abstract In this research, we introduce an innovative approach, for selecting genes in microarray-based cancer classification. Analysing gene expression using microarrays is crucial for disease and cancer detection. However, identifying the relevant gene markers is challenging due to the nature and high dimensional aspects of the data. We introduce the BCOOT (Binary COOT) optimization algorithm, which shows potential for gene selection tasks. We propose three variations; BCOOT, BCOOT-C and BCOOT-CGA. In our approach we transform the COOT algorithm into binary form using a hyperbolic tangent transfer function. The second strategy enhances exploration by incorporating a crossover operator (C) into BCOOT. For our method BCOOT-CGA we combine BCOOT C with a Genetic Algorithm to strengthen exploitation and identify robust and informative genes. To improve the gene selection process further we include a prefiltering step called redundancy relevance (mRMR) technique to eliminate redundant genes. To evaluate our proposed algorithms performance we conduct assessments, on six established microarray datasets comparing them with other robust optimization techniques and state of the art gene selection methodologies. In the classification step of our study, we utilize a Random Forest classifier. The experimental findings showcase that the BCOOT-CGA approach outperforms both BCOOT and BCOOT-C, surpassing alternative methods regarding the accuracy of predictions and the quantity of chosen genes in the majority of instances. This underscores the effectiveness of our proposed approach in enhancing microarray-based cancer classification, highlighting its potential to advance disease diagnosis and prognosis.