Integrated Classifier: A Tool for Microarray Analysis

Abstract

Microarray technology has been developed and applied in different biological context, especially for the purpose of monitoring the expression levels of thousands of genes simultaneously. In this regard, analysis of such data requires sophisticated computational tools. Hence, we confined ourselves to propose a tool for the analysis of microarray data. For this purpose, a feature selection scheme is integrated with the classical supervised classifiers like Support Vector Machine, K-Nearest Neighbor, Decision Tree and Naive Bayes, separately to improve the classification performance, named as Integrated Classifiers. Here feature selection scheme generates bootstrap samples that are used to create diverse and informative features using Principal Component Analysis. Thereafter, such features are multiplied with the original data in order create training and testing data for the classifiers. Final classification results are obtained on test data by computing posterior probability. The performance of the proposed integrated classifiers with respect to their conventional classifiers is demonstrated on 12 microarray datasets. The results show that the integrated classifiers boost the performance up to 25.90% for a dataset, while the average performance gain is 9.74%, over the conventional classifiers. The superiority of the results has also been established through statistical significance test.