Databases containing huge amount of genetic information about diseases related to cancer are beyond our capability to analyze and predict the discriminative characteristics of the genes involved. But, this kind of analysis helps to find the cause and subsequent treatment for any disease. In this work, a hybrid model has been developed combining the characteristics of Rough set theory (RST) and Correlation based feature subset (CFS) selection technique which is capable of identifying discriminative genes from the microarray dataset. The model is tested with two publicly available multi-category microarray dataset such as Lung and Leukemia cancer. The study reveals that Rough set theory (RST) is capable of extracting predictive genes in the form of reducts from the subset of genes which are highly correlated with the class but having low interaction with each other. The performance of the model has been evaluated via three learning algorithms using 10-fold cross validation. This experiment has established that the hybrid supervised correlation based reduct set (CFS-RST) method is able to identify the hidden relationships among the genes which cause diseases as well as help to automate medical diagnosis. Finally, the functions of identified genes are analysed and validated with gene ontology website DAVID which shows the relationship of genes with the disease.