Gene expression analysis is an efficient modality to detect breast cancer. But analysis of the gene expression data is a complicated task due to the involvement of large number of numerical data for each subject. This research work focuses on the development of computerized diagnosis systems for analyzing the gene expression data and classifying the subjects as either normal or breast adenocarcinoma. Sparse autoencoder is used for dimensionality reduction and Elephant Herding Optimization is used for transforming the features. Four different supervised classifiers namely support vector machine, random forest classifier, decision trees, and naive bayes classifier are examined for classification task. Among them, random forest provides the superior balanced accuracy of 86% while it was further enhanced to 92% with the help of sparse autoencoders and elephant herding optimization algorithm.