In the human body, where the greatest concentration of bacteria is the gastrointestinal tract, it is considered to be a diverse and complex microbial population, involving many different diseases. The development of metagenomics has many achievements in evolution and biodiversity. The application of machine learning algorithms to solve metagenomics problems has helped researchers make new advances in the field of personalized medicine, especially the diagnosis and improvement of human health people. In this study, we propose an unattended binning approach combined with Mean-shift algorithm to improve predictive performance. We performed on the Inflammatory Bowel Disease (IDB) dataset with 6 subclasses. This clustering method has improved results when applying deep learning techniques and shows the promising potential of data preprocessing methods when applied on different datasets.