Microarray Lung Cancer Data Classification Using Similarity Based Feature Selection

Abstract

Microarray data allows monitoring of thousands of genes simultaneously in one experience; arrays are applied to understand gene expression. Gene expression synthesizes proteins, which perform essential functions for all human cells. Features (genes) selection techniques aims to reduce the dimensionality of genes, it removes correlated, redundant and irrelevant data to find best target genes. Best genes are usually used as a biomarker to reduce experience time. In one microarray experiment we have more genes to study than instances. In this work we used five methods of similarity based techniques (Fisher, Relief, SPEC, Trace Ration, Laplacien)to find 2000 most relevant genes, then we classify our data with K-nearest neighbor, we found that fisher method gave better result with 3 and 5-nearest neighbor, it achieved an accuracy equals to 94.50% and 82.33.