A Combined Filter Wrapper Classification Method for Gene Selection from Gene Expression Datasets

Abstract

In the generic sense, Gene Selection methods are implemented upon a huge gene bank to decisively corner and expose certain genes that are indicative of say, diseases with their own set of classifications. The lightening surge about the DNA microarray dataset and its huge influence in the scientific realm has led different fields with the likes of Ecology, Bioinformatics, Computer Science, etc., making giant strides in their respective researches. DNA microarray research field threw open a desirable scope for path-breaking methods to be employed for gene selection, aimed at classifying those informative genes. Gene expression data classification is realized and aspired at the wake of a huge data size, boasting a usually miscellaneous yet a dissuasive composition that serves a challenge for data miners. The ideas and research work expressed below is a cohesive approach where a hybrid method linking every filter method (Information Gain / Pearson Correlation Coefficient / Relief-F) with that of wrapper (Genetic Algorithm / Forward Selection Backward Elimination / Practical Swam Optimization), through all permutation and combination, the accuracy of gene data (after being put through Support Vector Machine (SVM) model classifier) is optimized to the maximum and authenticated, yielding the optimum results in accordance with the requirements. Comparison between all filter, wrapper and hybrid methods are done by applying it on three microarray cancerous dataset.