The early detection of at-risk human diseases is a significant challenge for health care professors. Data Volumes related to the disease are usually relatively large, but the capability of human computation is limited. This study aims to provide a solution for disease prediction by selecting characteristics of the original set of features on metagenomic data. We propose a novel approach to enhance the prognosis of inflammatory bowel disease (IBD) and colorectal cancer (CRC), which are to use Random Forests and the Support Vector Machine (SVM). The results with the selected features using the proposed method are pretty promising on datasets of Colorectal Cancer and Inflammatory Bowel Disease (IBD) compared to the original set of features using state-of-the-art techniques.