基于电子病历分析的糖尿病患病风险数据挖掘方法研究

发布时间：2018-05-12 11:13

本文选题：空腹血糖 + 数据挖掘　；参考：《青岛大学》2016年硕士论文

【摘要】：空腹血糖(Fasting Blood Glucose,FBG)是衡量人体健康的一个重要指标。对空腹血糖的分析预测对发现和治疗疾病有重要意义,尤其是对糖尿病等相关疾病。体检数据是一种信息丰富的电子医疗数据,其中包含着大量体检者长期的健康指标。从大数据分析的角度,本文主要利用大数据方法研究了众多体检项与空腹血糖变化之间的关系,包含两方面的工作:首先,基于连续四年的历史体检数据和统的数据挖掘技术和经典算法,提出了对未来空腹血糖变化趋势预测的模型。该模型包括结合了随即森林特征重要性和顺序前向选择(Sequential Forward Selection,SFS)方法的特征选择算法,和对空腹血糖变化趋势概率化表示的算法。利用体检数据进行实验,表明进行特征选择后,得到与空腹血糖相关性较大体检项,并且模型的性能得到提升。对测试集上空腹血糖的趋势概率化表示,并选取空腹血糖变化高概率人群,从敏感度、特异度、正阳性预测值三方面进行分析,发现对于空腹血糖上升人群具有较好效果。其次,针对空腹血糖值表现正常群体,研究体检指标与空腹血糖高风险人群的关系建立空腹血糖上升风险预警模型。该模型考虑任两个体检项相互作用对于空腹血糖变化的影响,在数据集中添加了交叉项,通过重要体检项得分和特征选择,发现哪些体检项的相互作用对空腹血糖。从构造的交叉特征集合中,使用特征选择方式选取重要交叉项特征,用于构建空腹血糖上升风险预警模型。利用体检数据进行实验,结果表明随机森林模型在加入经过特征选择选取的交叉项,模型的性能得到提升,并对重要特征进行分析。由于数据集中存在严重的正负样本不平衡分类问题,本模型设计了基于欠抽样和模型组合的方法处理数据不平衡分类问题,取得了良好的效果。
[Abstract]:Fasting Blood Glucose (FBG) is an important indicator of human health. The analysis and prediction of fasting blood glucose is of great significance to the discovery and treatment of diseases, especially for diabetes and other diseases. Physical examination data is an informative electronic medical data, which includes a large number of long-term health indicators for a large number of medical examiners. In the perspective of large data analysis, this paper mainly uses large data method to study the relationship between many physical examination items and the change of fasting blood glucose. It includes two aspects: first, based on the historical physical examination data of four years and the data mining and classical algorithms, the model of predicting the trend of the change of fasting blood glucose in the future is put forward. The model includes the feature selection algorithm combined with the forthcoming forest feature importance and the sequential forward selection (Sequential Forward Selection, SFS) method, and the algorithm for the probability expression of the fasting blood glucose change trend. The performance of the type was improved. The trend of fasting blood glucose was expressed in the test set, and the high probability population of fasting blood glucose change was selected. From the sensitivity, specificity and positive positive predictive value three, it was found that it had a good effect on the population with ascending fasting blood glucose. Secondly, the normal group of fasting blood glucose was shown to study physical examination. The relationship between the index and the high risk population of fasting blood glucose was established to establish an early warning model for the risk of fasting blood glucose rise. The model considered the effect of the interaction of two physical examination items on fasting blood glucose change. In the cross feature set of the structure, the feature selection method is used to select the characteristics of the important cross item, which is used to construct the risk early warning model of the fasting blood glucose rise. The results show that the performance of the model is improved and the important characteristics are analyzed by adding the cross items selected by the feature selection. Because of the serious negative and negative sample classification problem in the data set, this model is designed to deal with the problem of disequilibrium classification of data based on the method of undersampling and model combination, and good results have been achieved.

【学位授予单位】：青岛大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：R587.1;TP311.13

【相似文献】