Abstract: Cardiovascular disease is a common disease threatening human health. In order to predict it more accurately, this paper optimizes and improves the traditional DNN model and proposes a directional regular deep neural network (TR-DNN) model. By improving the defects of the original deep neural network model, it can better train and test the cardiovascular disease data set, further realize the task of cardiovascular disease prediction. Experiments show that the model performs well in data set training, and achieves excellent results in test set. Finally, comparing the results of TR-DNN with SVM, RF and XGBoost models in the same data set, the evaluation indexes of TR-DNN model are better than other models. Compared with the traditional DNN model, TR-DNN model improves the accuracy by 1.507 percentage points, the recall by 1.57 percentage points, the specificity by 2.54 percentage points and the precision by 1.51 percentage points. Therefore, TR-DNN model can be applied to the prediction of cardiovascular disease.

Key words: prediction of cardiovascular disease, DNN, auxiliary diagnosis, optimization algorithm