Prediction of Hypertension Using Deep Autoencoder-Based Feature Representation

As the elderly population increases, the number of patients with chronic diseases, which usually occur in the elderly population, continues to increase. Many studies have been conducted to predict chronic disease using various medical data. Chronic diseases are mainly caused by complex factors rather than independent factors. In this study, the disease was predicted by considering KNHNAES data and air pollution as complex factors. However, when considering complex factors, the accuracy of disease detection is poor because high dimensions require high computational complexity. In order to overcome this high-dimensional problem many studies have been carried out feature representation method. The feature representation approach plays an important role in the success of classification, while ex-pressing high-dimensional features in a low dimension. This study used deep autoencoder to compress representative feature by applying variational autoencoder. In the experiment, the proposed feature representation method and conventional feature reduction method were compared. The results showed that the proposed method outperformed the conventional method. The compressed representative features to predict hypertension using xgboost method showed best performance.