Research On Coronary Heart Disease Screening Model Based On Ensemble Features Selection

Cardiovascular disease is a serious threat to human health.Coronary heart disease(CHD),as a common cardiovascular disease,its morbidity and mortality are increasing year by year.At the same time,its expensive treatment costs have brought huge economic burden to the country and people,and great disaster to the patients'families.Coronary angiography is a commonly used invasive diagnostic technique in recent years,which is considered as the "gold standard" for the diagnosis of CHD.Therefore,it is of great significance to develop a universal early non-destructive screening diagnosis method for CHD to reduce the morbidity and mortality of CHD.In this paper,the research status and development trend of CHD risk factors and screening models at home and abroad were reviewed,and then the CHD risk factors were screened by multiple feature selection methods,and the establishment of early nondestructive screening models for CHD was systematically studied.The main innovative work completed is as follows:(1)complete the data collection of CHD group and control group in the department of cardiology of a grade a hospital and build the database of CHD risk factors in Shandong.The main information included clinical symptoms,biochemical indexes,demographic information,life habits,personal disease history,family disease history,electrocardiogram results and so on.(2)Anew feature selection strategy integrating multiple feature selection methods is proposed.First of all,using analysis of variance,chi-square,mutual information,circulation recursively eliminate,random forest features weight coefficient,support vector machine weight coefficient and XGB(Extreme Gradient Boosting,XGB)feature weighting coefficient of seven kinds of feature selection methods to complete assessment of importance of the above features set and selected important feature;Then,for the selected feature votes,the votes of each feature are counted,and the features with the same number of votes constitute a new feature set;Finally,the new features were modeled using support vector mechanisms,and the most important feature sets were obtained through the model performance indexes.(3)The CHD screening model was established based on the support vector mechanism and the most important feature.The accuracy,sensitivity and fl-measure of the model reached 83.88%,88.92%and 88.73%,respectively.Compared with the model trained by the original feature set and the model trained by the single feature selection method,the accuracy of the model trained by the integrated feature selection method is improved by 23%and 4%,respectively.The model trained by this method can effectively identify CHD and provide reference for the screening and diagnosis of clinical CHD.