Research On Key Techniques Of Cardiovascular Diseases Risk Prediction Based On Machine Learning

Machine learning is a key technique in artificial intelligence research and has a wide application foreground in medical and health.Cardiovascular diseases(CVD)is one of the most serious health threats to human society in the world.Using machine learning method to correctly forecast CVD can effectively identify high-risk patients and risk factors,timely take appropriate intervention operations for patients and help medical institutions optimize resource allocation,which has important practical significance.Aiming at the risk prediction of CVD,this paper carried out the following aspects of research:(1)We studied the prediction method of hospital admissions for CVD based on machine learning models.In terms of data collection,environmental factors were also included in the study in addition to the basic data of hospital admissions.Then,the direct prediction strategy was adopted on machine learning models to construct forecast models of hospital admissions for CVD,and the results were compared with the traditional LR model.Finally,feature importance analysis is carried out.The results suggested that machine learning models were better than LR model in each evaluation metric,and the Adaboost model had the best prediction performance.(2)We studied the risk prediction approaches for hospital readmissions of patients with Acute Myocardial Infarction(AMI),and proposed an ensemble learning model based on Stacking technique to forecast the risk of 30-day unplanned hospital readmissions for AMI patients.We used an adaptive strategy to select some models from multiple candidate models as base classifiers,and built a three-layer Stacking ensemble learning model on theses base classifiers.Then,we compared the prediction results between the Stacking model and machine learning models.The experimental results suggested that,compared with machine learning classifiers,the proposed Stacking model achieved the best forecast performance in each evaluation metric.(3)We studied AMI in-hospital mortality risk prediction on the basis of deep learning.First of all,we built the multi-layer perceptron(MLP)model in deep learning.Then we used a filtering feature selection method which could combine the advantages of linear model and tree model for feature selection.Finally,we used layer-wise relevance propagation(LRP)method to study the interpretability.The experimental results showed that the MLP model established in this research had the best forecast performance compared with other models.The filtering feature selection method could greatly improve the prediction performance and reduce the running time of the model,and the LRP method could also have a good interpretation effect on clinical structured data.