Generative Adversarial Networks For Heart Disease Prediction

Heart disease often poses a great threat to human health for the particular location of heart and the particularity of myocardial cells.Using the information of medical indicators related to heart disease to predict the state of heart disease can provide people with risk reference and buy time for prevention and treatment with less medical resources.At present,the prediction of heart disease mainly focuses on the use of machine learning classification algorithms such as support vector machine,and other measured medical indicators such as exercise angina pectoris are used as input to predict the risk of heart disease.By adjusting the parameters of the algorithm,a better prediction effect can be achieved.The training effect of machine learning algorithm is not only influenced by the model itself,but also by the training data.However,the medical data related to heart disease prediction used in most studies are relatively small in sample size.The training data set should be rich enough to make the trained model have good robustness.Lack of data will lead to the loss of some information and adversely affect the training process of related models.The adversarial generative network can fully capture the information such as probability distribution contained in the original data,and generate synthetic data with similar distribution,by using the special training strategy that make the generator and the discriminator contes.In this paper,TGAN,a adversarial generative network,is applied to heart disease prediction to learn the probability distribution of the original data,which can improve the training effect of the classification algorithm by enhancing the training data.TGAN is composed of generator,short-term and long-term memory network,and discriminator,multilayer perceptron.When training reaches Nash equilibrium,synthetic data with similar distribution to the original data can be generated and added to the original data to enhance the training data.On this basis,the classification algorithm is trained.The results show that the prediction accuracy of decision tree,random forest and support vector machine is improved when synthetic data is added to the training set,compared with that when only raw data is used for training.On AUC and F1-Socre,which describe the performance of the algorithm,the algorithm on the training set containing synthetic data performs better than the original data.The above results show that the training effect of the classification model can be effectively improved by generating high-quality synthetic data to enhance the training data set.This paper also discusses the influence of the amount of synthetic data added on the training effect of the algorithm.Although this paper takes heart disease data as the research object,this attempt shows the ability of adversarial generative network to enhanc data,which can be similarly applied to other problems with data missing,so as to improve the data quality of the training set.