Chi Square and Support Vector Machine with Recursive Feature Elimination for Gene Expression Data Classification

Abstract

As a result of the rapid evolution of microarray technology, plenty of statistical research that aims to detect the various expressed genes has been raised. Many studies were conducted on DNA microarray data. At present, there are several methods used in analyzing DNA microarray to help in dealing with various related research. Microarray generates large and complex data with high dimensionality issues. Thereby, the curse of dimensionality may reduce the effectiveness and performance of classifications and increase computational complexity. Therefore, feature selection techniques work to solve dimensionality issues by choosing informative genes. In this study, we proposed a new combination of feature selection methods called ChiSVMRFE based on the Chi Square Statistic and Support vector machine with Recursive Feature Elimination SVMRFE. Chi-Square used as a ranking method to calculate the weight of genes with respect to the class label, then the top ten percent of genes with the higher weights were considered as relevant and important genes. SVMRFE repeatedly train a model to discard features with the lowest weights. Finally, SVMRFE selects ten features to consider informative genes. The proposed method was applied to eleven microarray high dimensional datasets. The ChiSVMRFE worked effectively to select only ten genes that considered informative genes, also, showed improvement in the classification results compared with other methods in previous studies.