Feature Selection from Microarray Data based on Deep Learning Approach

In the field of medical care, feature (gene) selection has been a chronic research topic. For the feature selection, microarray data are an exigent part to diagnose any disease. In any microarray data, there are more than thousands of features. Therefore, it becomes an important object to find out a proper subset of features with the help of conventional algorithms. To generate a proper subset, we have to reduce the dimension of the whole dataset after removing some redundant features without missing significant features. Over and above, small features set with minimum number of samples can more usefully to prognosis any disease. Autoencoder technique is a powerful technique to reduce the dimension. Inspired by this, in this paper, we built a model based on Folded Autoencoder (FA) to select a feature set. After that, we have applied some machine learning classifiers to check the classification accuracy. Among all the classifiers, Support Vector Machine (SVM) gives better result after reducing the dimension of features. We named this model as Folded Autoencoder-SVM (FAS). We have compared the result among whole dataset (without applying FA) and reduced dataset (after applying FA).