A hybrid multi-objective genetic algorithm for gene selection in microarray data

The classification of microarray data is crucial for cancer diagnosis and prognosis. However, the high dimensionality of gene microarray data often leads to suboptimal classification performance. Furthermore, only a small subset of genes in vast datasets significantly contributes to accurate disease classification, emphasizing the importance of feature selection in this domain. This study introduces a novel hybrid feature selection method, denoted as the Maximum Relevance Non-Dominated Sorting Genetic Algorithm (MRNSGA). This approach makes use of gene-gene correlations and redundancies to facilitate the initialization of the genetic algorithm population. Additionally, a mutation retry operator is incorporated into the genetic algorithm to enhance its performance. The proposed method is compared with other advanced evolutionary algorithms across 15 gene microarray datasets. The results show that our algorithm significantly improves gene classification accuracy and considerably reduces the time required for feature selection.