Research On Stroke Recurrence Prediction Based On Machine Learning

Stroke is also called the cerebral apoplexy, is a kind of clinical common disease and frequently-occurring disease, seriously threatens people’s health. It is also known as a cerebrovascular accident, disease which caused by various factors to induce the brain artery stenosis, occlusion, or this breakdown phenomenon. First day after the onset of cerebral apoplexy, about 10% ~ 20% of patients died, the onset of about 1/4 ~1/3 of patients died within three weeks, and after 3 weeks the death rate will slow down, after 5 year the survival rate is about 15% ~ 40%, and those who had stroke recurring again have 4 ~ 15 times that of ordinary people, and the treatment will be very difficult.Stroke has high incidence, high morbidity, high mortality and high rate of relapse of characteristics. Therefore, it is very important to prevent recurrence of stroke, to reduce morbidity, improve the cure rate. While most people will appear stroke sign before stroke occurs, therefore, in apoplexy disease prevention aspect, this paper mainly research content is as follows:Firstly, for the topic background of cardiovascular disease, this paper discusses the etiology and pathogenesis of heart head blood-vessel, at the same time, in detail elaborated the cardiovascular disease recurrence forecast model research status both at home and abroad, and introduces the common cardiovascular disease assessment tools: total CVD risk prediction, stroke prediction tools and cerebral hemodynamic measurements;Secondly, this paper introduces a risk factor for cardiovascular disease, such as gender, age, family history, hypertension, smoking and other factors. The paper expounds the concept and classification of machine learning. The paper introduces the use of SPSS software and its function features.Then, according to the characteristics of cardiovascular disease, this paper establishes the prediction model of cerebral apoplexy based on Cox regression, and through SPSS13.0 software to test, then the corresponding experimental results are given; At the same time the paper sets up a forecasting model based on Logistic,carries on the related experiment and result analysis.Finally, the paper designs a prediction model based on machine learning methods,expounds its performance indicators: the area under the ROC curve, and introduces indetail the specific steps of the algorithm:(1) the import related database;(2) feature selection;(3) using machine learning algorithms to evaluate forecasting performance.It gives the related experimental results and analysis. And at the same time, compared with the Cox model and logistic model, it finds that the SVM combined CM feature selection algorithm has the best effect.