Variational Autoencoder-Based Deep Neural Network for Coronary Heart Disease Risk Prediction

Tsatsral Amarbayasgalan, Kwang Ho Park, Khishigsuren Davagdorj, Keun Ho Ryu, Nipon Theera-Umpon

Abstract: Coronary heart disease (CHD) is one of the top causes of global mortality. Most patients cannot be diagnosed at the early stage because it does not give any symptoms for many years. If CHD gets worse, it will require advanced treatments, such as heart transplant and stent surgery. Therefore, it is useful in preventing CHD by predicting high-risk people who will suffer from CHD. In this study, we have proposed a variational autoencoder (VAE)-based deep neural network (DNN) model for predicting CHD risk. We improved the performance of DNN model by enriching training dataset using VAE-based data generation. The proposed method has been compared with machine learning algorithms on Korea National Health and Nutrition Examination Survey (KNHANES) dataset. As a result, the proposed method outperformed the compared classifiers. The performance measurements include accuracy, precision, recall, F1-score, and AUC score which reached 0.851, 0.882, 0.809%, 0.844, and 0.851, respectively.

External IDs:doi:10.1007/978-981-19-1057-9_1