Disease Single Nucleotide Polymorphism Selection using Hybrid Feature Selection Technique

Abstract

According to recent studies the Single Nucleotide Polymorphism (SNPs) plays very important role as genetic marker in various complex diseases. Lots of machine learning techniques are already applied on SNPs data to distinguish between affected and healthy individuals. The major problem with the SNPs dataset is high number of features and small number of samples which are referred as 'large p' and 'small s' problem. In this paper we proposed a hybrid feature selection method for selecting an optimal subset of SNPs and from that we select the significant SNPs, which act as marker for disease. The method is a hybrid technique based on combination of filter and wrapper method, the (mRMR) Minimum Redundancy Maximum Relevancy and Particle Swarm Optimization for Gene Selection with Support Vector machine (PGOGS-SVM) respectively. The proposed mRMR+PSOGS-SVM approach has been applied to mental retardation SNP dataset taken from NCBI-GEO website. The method has achieved high classification accuracy up to 88% and outperformed all other compared feature selection techniques.