The Next Generation Sequencing technologies provide large volumes of DNA-seq and RNA-seq data. A central part of their investigation is the task for selecting the differentially expressed genes. Different methods for RNA-seq data analysis that identify genes distinguished by their expression levels have been proposed basically on the statistical data analysis. There is no agreement among the applied methods as different results are produced by the distinct methods. The present paper proposes a new method for differential gene expression analysis based on machine learning approach. Difficulty of the selection due to the large number of indistinguishable genes is solved by iterative clustering procedure. The importance of the proper cluster distance measure is discussed. The visibility of the procedure results and ability to find different number of compact clusters is also underlined. The significance of the method is investigated and proved by application to the two mice strains dataset. The obtained results are compared with the results of the statistical methods applied to the same dataset. It is concluded that the proposed method is valuable and could be applied as standalone or for preliminary genes selection within a statistical algorithms pipeline for discovering differentially expressed genes.