基于决策树、逻辑回归和改进神经网络的几种慢性病的危险因素分析研究

摘要

Analyzing the risk factors of hypertension, diabetes, hyperlipidemia and coronary Heart Disease(CHD) with several data mining methods, such as decision tree, logistic regression(LR) and an improved Neural network based on gra-dient descent (GDNN). Then further research the correlation among those risk factors and find their common risk factors include AGE, SEX, SBP, TG, TC, BMI, 2hPPG, LDL-C, Smoking, diabetes, Hyperlimpedia, CHD, Hypertension. And then we discovery that C4.5, LR, GDNN, BNN fit for mining risk factors of Hypertension; C4.5, GDNN fit for mining risk fac-tors of Diabetes, Hyperlipidemia and CHD. Finally, we also find that the GDNN outperforms than BNN in this paper.%本文采用决策树(C4.5)、逻辑回归(LR)和一种改进的神经网络(基于梯度下降, GDNN),分析高血压、高血脂、糖尿病、冠心病这四种慢性病各自的危险因素和共同的危险因素,以观察四种慢性病之间的关联关系。通过本文的研究表明：(1)四种慢性病共同的危险因素有：年龄、性别、收缩压、甘油三酯、总胆固醇、BMI、餐后2h 血糖、LDL-C、吸烟、糖尿病、高血脂、冠心病、高血压；(2)我们还发现 C4.5、LR、GDNN、BNN 都适用于与分析高血压的危险因素；(3)C4.5, GDNN比 LR 和 BNN 更适用于分析糖尿病、高血脂、冠心病的危险因素；(4)GDNN 在分析四种慢性的危险因素时,其准确度高于BNN。