標題: 使用數學建模篩選糖尿病的高風險群
Identify High Risk Group of Diabetes Mellitus Using Mathematical Modeling
作者: 蔡毓珊
何信瑩
Cai,Yu-Shan
Ho,Shinn-Ying
生物資訊及系統生物研究所
關鍵字: 糖尿病;機器學習;繼承式雙目標基因演算法;演化篩選演算法;國民健康訪問調查資料庫;Diabetes mellitus;Mathematical Modeling;Evolutionary screening algorithm(ESA);National Health Interview Survey Database(NHIS);Inheritable Bi-objective Genetic Algorithm(IBCGA)
公開日期: 2017
摘要: 背景:糖尿病初級預防現今研究多透過非侵入式的檢驗來篩選糖尿病的高風險群。因此本研究期望透過問卷的方式來了解生活型態,並且挑選一組重要的風險因子進行預測糖尿病,並達到糖尿病個人初級預防。 方法:本研究將使用台灣國家衛生研究院所提供之全民健康訪問調查資料庫來進行國人生活飲食型態與糖尿病相關影響的研究。罹患糖尿病樣本為866人,未驗證健康人群為2595人,利用Evolutionary screening algorithm(ESA) 篩選出有效的健康樣本,並利用繼承式雙目標基因演算法(Inheritable Bi-objective Genetic Algorithm, IBCGA)挑選出重要的風險因子,最後並利用Support Vector Machine(SVM)進行數學建模。 結果:經過篩選出有效的健康樣本且特徵值挑選後,挑選共 30 個重要風險因子,包括家族遺傳史(糖尿病、心臟病、癌症)、是否有服用降血壓的藥物、視力問題、牙齒狀況、就醫狀況、血壓狀況、血脂狀況、飲酒、坐著的時間、飲食狀況、心理狀況、性別、工作狀況,在測試資料中敏感性提高了20%達到91%。 結論:研究結果表示經過有效的健康樣本篩選後,會提升數學模型預測有糖尿病的準確率之能力,最後找到30個重要風險因子可預測糖尿病的機率,使國民能夠初級預防糖尿病的發生。
Background: The Prevention of primary Diabetes always screens high risk group of Diabetes through the non-invasive of inspection. Therefore, I look forward to comprehend the lifestyle in this research and pick an impact factor which is more important to predict the Diabetes mellitus. So it will achieve individual primary prevention of diabetes. Methods: It will use the Databank of National Health Interview Survey Database to research the Healthy Eating Style of Taiwanese and Diabetes mellitus. After data processing, there has 886 samples suffer from a Diabetes, 2595 unverified healthy samples. To screen useful healthy samples by using Evolutionary screening algorithm(ESA) and choose an importance of impact factor. At last use the Support Vector Machine(SVM) to build the Mathematical Modeling. Results: Totally choosing 30 impact factor after the feature selection, including household heredity factors (Diabetes, hyperlipidemia, heart disease),taking blood pressure lowering drugs, vision problems, dental condition, medical condition, blood pressure status, blood lipid status, Drinking, sitting time, diet status, psychological status, gender, working condition. Also use this Mathematical Modeling to predict it, and it increased by 20% to 91% in the sensitivity of testing data. Conclusions: The results of the study show that the health samples will enhance the ability of the mathematical model to predict the accuracy of diabetes. Finally found 30 important factors to predict the Diabetes mellitus. And it also can be used to prevent the Primary of Diabetes mellitus.
URI: http://etd.lib.nctu.edu.tw/cdrfb3/record/nctu/#GT070457049
http://hdl.handle.net/11536/141867
Appears in Collections:Thesis