Hybridising harmony search with a Markov blanket for gene selection problems

Gene selection, which is a well-known NP-hard problem, is a challenging task that has been the subject of a large amount of research, especially in relation to classification tasks. This problem addresses the identification of the smallest possible set of genes that could achieve good predictive performance. Many gene selection algorithms have been proposed; however, because the search space increases exponentially with the number of genes, finding the best possible approach for a solution that would limit the search space is crucial. Metaheuristic approaches have the ability to discover a promising area without exploring the whole solution space. Hence, we propose a new method that hybridises the Harmony Search Algorithm (HSA) and the Markov Blanket (MB), called HSA-MB, for gene selection in classification problems. In this proposed approach, the HSA (as a wrapper approach) improvises a new harmony that is passed to the MB (treated as a filter approach) for further improvement. The addition and deletion of operators based on gene ranking information is used in the MB algorithm to further improve the harmony and to fine-tune the search space. The HSA-MB algorithm method works especially well on selected genes with higher correlation coefficients based on symmetrical uncertainty. Ten microarray datasets were experimented on, and the results demonstrate that the HSA-MB has a performance that is comparable to state-of-the-art approaches. HSA-MB yields very small sets of genes while preserving the classification accuracy. The results suggest that HSA-MB has a high potential for being an alternative method of gene selection when applied to microarray data and can be of benefit in clinical practice.