In the modern field of genomics there has been a lot of immense awareness in research, due to the formation of new and advanced data banks. These data banks act as a centre of all medical research biological views, which are functional and actively updated by various boards of bio-medical research council (NCBI). These data warehouses host a lot of multi-dimensional databases readily available for active analysis, algorithmic problem solving and efficient optimization to gain higher information. In this paper, it uses gene disease datasets to be acted on by differential gene analysis approaches such as Log2mean and negative binomial model to extract the most differential genes. Thus obtained genes undergo a feature transformation approachsuch as PCA to obtain highly significant genes. Then the data sets are learned using unsupervised learning based classification approach by using SOM based classifier to classify samples to give an efficient accuracy in predicting the categorical disease type samples. Thus this gives a comparison analysis of classification with feature transformation and classification without feature transformation. Similarly feature selection based classification is compared with transformed feature based classification. It also uses a hybrid approach of using feature transformation on selected genes and the categorical class samples while classification gives efficient accuracy on classification. The classification onL2M-NBIN based PCA transformed features shows better efficient accuracy than normal transformed PCA genes and selected L2M-NBIN based classification approaches.