| Summary: | 碩士 === 國立屏東科技大學 === 資訊管理系所 === 103 === Along with socio-economic prosperity and lifestyle change, global prevalence of diabetes has been increasing every year. Diabetes also becomes one of the ten leading causes of death in our country. The complications include foot neuropathy which may need amputation, and retinopathy which causes blindness and renal failure. The risk of heart disease and stroke also elevates. If the high-risk group of diabetes can be discovered early and be provided with appropriate diet and exercise suggestions early, we can avoid the patients suffering from diabetes.
This study tried to use data mining techniques to study the report data of adult health exams in a regional hospital. The main purpose was to develop a set of pre-diabetes predictive model, and to discuss the risk factors of pre-diabetes. The research samples were extracted through simple K-means clustering algorithm, and then were analyzed by six classifiers (Decision trees, Logistic Regression, Multilayer perceptron, SMO, Naive Bayes and RBF Network). We compared these six prediction modes in order to find the best one. The research results showed: (1) the prediction mode analyzed by Logistic Regression had the best performance, and the accuracy was 99.8%, the sensitivity was 98.7%, the specificity was 99.9%, and the positive predictive value (PPV) was 99.1%; (2) this study found five risk factors of pre-diabetes, including systolic blood pressure, diastolic blood pressure, body mass index (BMI), cholesterol level and fasting blood glucose level; (3) with decision trees, this study arranged 11 rules which are practically useful and easily interpretable. Before diabetes develops, the patients should be provided with medical personnel diagnostic aids or individual disease prevention, in order to discover the disease early, receive proper diet and exercise suggestions early as well as reduce the risk factors, so that the patients will not suffer from diabetes.
Keyword：Pre-diabetes, Predictive model, Risk factor, Data mining
|
|---|
