Application Study Of SVM In Prediction Of Coronary Heart Disease

Cardiovascular disease (CVD) has been recognized as the dominant cause of death formany years in developed countries. High incidence of cardiovascular disease with the highmorbidity rate and causing paralysis is its main charateristic, and its high cost of treatment hasbrought huge economic burden to society and the country. Cardiovascular disease also willbring great distaster to the family if any member suffers from it. Early detection and earlyintervention in cardiovascular disease patients will not only avoid the loss of ability to work,but also reducing the subsequent huge medical expenses at the same time. It is very importantto make a research on cardiovascular disease, build a prediction model of cardiovasculardisease, and it will make a gteat contribution for cardiovascular disease detection andtreatment. In this study, southern Chinese people as research subject,1237cases of sampleswere obtained, including620positive cases (i.e. coronary heart disease patients), and617negative cases (i.e. non-coronary heart disease persons), and support vector machineprediction model of cardiovascular disease will be built. The main content includes thefollowing aspects:(1)The different physiological parameters between coronary heart disease patients andnon-coronary heart disease persons are analysised, and these sample are preprocessing forbuilding prediction model. The result shows that CHD patients’ mean physiologicalparameters are larger than normal individuals’ except for HDL-C, which is consistent withprevious reports. There is a lot of overlaps between positive samples and negative samples, sothat it is impossible discriminating coronary heart disease patients from non-coronary heartdisease presons with physiological parameters.(2)Support vector machines prediction model are built with these sample fordiscriminating coronary heart disease patients from non-coronary heart disease presons.According to previous reports, risk factors of CVD are not dependent usually, there are somerevelance between the nine physiological parameters that we pick out on the contrary. SVM isacknowledged to be a typical supervised learning model in machine learning, has beenwidely used in classification and regression analysis of high dimensional data. The resultswhich we obtain shows that SVM prediction performs excellent. At the same time, SVMparameters are optimized by grid optimization, and they are compared to SVM with randomparameters.(3)In this study, the SVM’s parameters, penalty factor C and kernel parameter, wereoptimized by particle swarm optimization to improve its classification accuracy, sensitivity, specificity. Kernel function and its parameters setting--penalty factor C and kernel parameter—significantly influence the classification accuracy during the SVM training procedure, thebest parameters should be searched in the solution space. In this sudy, the best optimazation isthe radial basis kernel SVM.(4)Other classification algorithms are compared to SVM and SVM optimized byparticle swarm optimization(PSO-SVM) with their classification accuracy. In this study, backpropagation neural network, linear discriminant analysis and logistic regression willcompared to SVM and PSO-SVM. Radial basis kernel SVM and polynomial kernel SVMoptimized by PSO performs best in our classification results.