Quantum-Inspired Owl Search Algorithm with Ensembles of Filter Methods for Gene Subset Selection from Microarray Data

Finding the optimum subset of genes for microarray classification is laborious because microarray data are often high-dimensional and contain many irrelevant and redundant genes. To overcome this problem, we have proposed a two-step technique. In the first step, to reduce the vast number of genes or features, an ensemble of popular rank-based feature selection algorithms with filter evaluation metrics are used to select a group of top-ranking genes. In the next step, the quantum-inspired owl search algorithm (QIOSAf), a new filter fitness function-based metaheuristic search technique incorporating concepts from quantum computing, is developed to identify the best subset of genes from the predetermined list. The experimental findings reveal that the ensemble approach in the first step can select more dominant groups of genes than each of the individual filters. Furthermore, it has been found that QIOSAf can reduce the cardinality of the selected optimum gene subset with comparable classification accuracy and requires lesser computational time than our earlier proposed QIOSA-based wrapper approach (i.e. QIOSAw). Besides, compared with three popular evolutionary feature subset selection algorithms, QIOSAf efficiently reduces the optimum cardinality of the gene subset while maintaining acceptable classification accuracy.