Abstract

In the bioinformatics field, microarray data analysis has achieved considerable interest for disease diagnosis. There are two important issues faced by algorithms of microarray data analysis that involves the excess number of genes in comparison with a considerably lower number of samples. Even though several algorithms and approaches exists for such high dimensional data, but such large search space with more of irrelevant genes deteriorates the performance of classifiers. These irrelevant genes not just complicate the learning algorithm but in addition, the learning algorithm provided with irrelevant genes is susceptible to over-fitting. Many of the available approaches use two-phase processes: feature selection, which is then followed by classification. This research work also analyses the details of these two steps for the problem of Leukemia cancer classification. This research work provides a review on few important feature selection methods used in microarray data and highlights the advantages and drawbacks of different techniques. Secondly, the important challenges faced by the traditional classifiers are resolved with the help of parallel classifiers. The significance and relevance of every feature selection and classification techniques are experimented with the help of a Leukemia dataset. These algorithms are implemented with success on MATLAB environment and comparative analysis is performed with Leukemia microarray gene expression data.

Paper Details