Noninvasive sensing of biological information is a developing research field for easy physical examination. In this work we extracted features of Non-alcoholic steatohepatitis, liver cancer, and periodontal disease by applying a machine learning technique to breath gas components, which were detected by the instrument we developed. Dataset labeled as Non-alcoholic steatohepatitis, liver cancer, periodontal disease, and healthy were learned by a Support Vector Machine algorithm. Leave-one-out cross validation of the subjects was performed for each of all the combinations of the gas components. By evaluating accuracy of classification ability for the left-out subject, we found that specific gas components can be significant biomarkers for these diseases.