Development of an Efficient Data Mining Classifier with Microarray Data Set for Gene Selection and Classification

Abstract

Microarray sample classification has been studied extensively using classification techniques in machine learning and pattern recognition. In a microarray chip, the number of genes available is far greater than that of samples, which is a serious problem and the gene expression reduction, is important one. Prior to sample classification, it is important to perform gene selection and more interpretable genes to be identified as biomarkers, so that a more efficient, accurate, and reliable performance in classification can be achieved. For this purpose, a hybrid scheme was proposed and this scheme is called as Single Filter � Single Wrapper Classifier (SFSW). In this technique, the Filter approach is used to select the data sets from the large Microarray and the Wrapper approach is used to classify the gene expressions from the selected data sets. From the available statistical report, it is revealed that the Filter method has a fast dimensionality reduction step for selecting a small set of genes at the cost of accuracy and the Wrapper method used for improving classification accuracy on this small set of genes at the cost of computational delay. However this approach also has some problems such as different filtered subset leads to complex evaluation. Hence an Efficient Hybrid Classifier called �Multiple-Filter-Multiple-Wrapper Technique� has been proposed with improved performance of SFSW. The use of Multiple Filters with different filter metrics ensures that useful biomarkers are unlikely to be screened out in the initial filter stage. The use of Multiple Wrappers is intended to optimize the reliability of the classification by establishing consensus among several classifiers. However, in this research work, we have identified that the Classification Accuracy of MFMW is poor due to the consideration of Indecisive Prediction Status. Hence, this Research Work is introduced an efficient classifier called ICS4-MFMW, which is focusing both the dimensionality reduction and Indecisive Status. This work is demonstrated its efficiency in terms of classification accuracy with Sensitivity and Specificity. From the result, it is revealed that our proposed work perform better as compared with the existing MFMW Classifier.