A novel filter approach for efficient selection and small round blue-cell tumor cancer detection using microarray gene expression data

Feature selection is a vital task in machine learning and data mining to reduce the dimensionality of the data and also improves the classification performance of an algorithm in terms of high precision, low computational cost, and low vulnerability. Various many technologies have been successfully applied in the previous experimental studies for tumor detection. The foremost challenging task of gene selection method is extracting informative genes contribution in the classification from the DNA microarray datasets with lesser computational load. In this paper, we propose the conglomeration of the Kendall Correlation (KC) and Filter based Feature Selection (FS) method for better classification and prediction. We demonstrate the extensive comparison of the effect of Kendall Correlation with FS methods, using Relief-F, Joint Mutual Information (JMI), and Mutual Information based feature selection (MIFS), Conditional mutual information maximization (CMIM), and Max-relevance & Min-Redundancy (mRMR). To measure the classification performance of four diverse supervised classifiers K-Nearest Neighbor (KNN), Support Vector Machines (SVM), Naïve Bayes (NB), and Decision Tree (DT) have been used on small round blue cell tumors (SRBCT) dataset. The result demonstrates that Kendall Correlation in accumulation with mRMR performances better than other combinations.