The Assessment Function of Diabetes Mellitus Risk using Oversampling and Ensemble Techniques

• Main Authors: Kai-Ru Jeng, 鄭凱儒
• Other Authors: 喻石生
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/cgi-bin/gs32/gsweb.cgi/login?o=dnclcdr&s=id=%22107NCHU5394016%22.&searchmode=basic

碩士 === 國立中興大學 === 資訊科學與工程學系所 === 107 === 　　Diabetes mellitus (DM) is metabolic abnormalities of disease. Patients will characterized by high blood glucose levels over a prolonged period. The symptoms of diabetes include sense of hunger, frequent urination and increased thirst. Diabetes is easily to be ignored because of lacking serious symptoms, which leads to treatment delayed by doctor. Without early control, diabetes may result in irreversible damages and makes quality of life down. Since the proportion of diabetes patients grows with increased standard of living, it has become the burden to many countries. It is more important to establish diabetes risk assessment model. 　　The glucose tolerance test is the standard method for detecting diabetes which is the most accurate. However, prevention is better than cure, we are finding some methods to detect even before pre-diabetes. The other methods are classical paper-and-pencil based risk assessment questionnaires and their online versions, by adding up the points scored for each answer like weight, height, family history, to point that if the patient is suffering from the diabetes risk. However, the accuracy is not very good. This article proposed an evaluation model for diabetes from mRNA feature values by using many different oversampling methods to balance the dataset and using many different machine learning algorithms to ensemble the final result. The result could help the doctors to make diagnoses and help the patient to control the condition early, which reduce the cost of the human resources and the loss of life quality.