Performance Study of Cancer Selection/Classification Algorithms Based on Microarray Data

Abstract

Microarray data has an important role in detecting and classifying all types of cancer tissues. In cancer researches, relatively low number of samples in microarray has always caused some problems in designing classifiers. So, microarray data is preprocessed through gene selection techniques and the genes which contain no information is discarded. Basically, a proper gene selection method can effectively improve the efficiency of diseases (cancers) classification. The purpose of this article is to compare different extraction algorithms of informative genes and also their different classification algorithms. First, ReliefF algorithms, information gain and normalized mutual information are introduced as algorithms used in order to extract feature and their features are noted. Then three classification algorithms, two proposed Bayesian Linear Discriminate Analysis (BLDA), Modified Support Vector Machine (υ-support Vector Machine) algorithms and Probabilistic Neural Network are compared in terms of classification accuracy. Implementation results show that combinational algorithm of normalized mutual information and BLDA classifier has best performance among other raised methods. So that, with applying this algorithm, classification accuracy in blood cancer data base is 95.34 percent.

Keywords

• DNA (deoxyribonucleic acid)
• Microarray
• Support Vector Machine
• Gene