Machine learning to promote health management through lifestyle changes for hypertension patients

The purpose of this paper is to investigate the use of machine learning models to develop a diagnostic system for hypertension patients so that people can modify their daily lifestyle to manage their condition. We propose this system by adopting the concepts of saliency maps for image data to non-image, lifestyle data with a data perturbation simulation technique. We trained the proposed system on a new lifestyle dataset that we extracted from a survey on Asian sub-population. The proposed system consists of a convolution neural network (CNN) as the diagnostic model, and is combined with simulation techniques to explain the concepts/insights learnt by the CNN. We compared classification performance of the CNN model with other baseline models fitted with other types of hypertension data including neural network, decision tree and other CNN model from literature. The CNN achieved a 68–70% accuracy on training and testing datasets. Comparing with other baseline models, our CNN model provided more consistent performance in terms of accuracy, sensitivity, specificity and area under receiver operating characteristic (ROC) curve. Using the simulations, we learnt that CNN captured not only direct correlation between the variables and the target, but also learnt group-based interactions. Our study reveals that age, gender, diabetes status, body mass index, smoking, occupation and education are some important lifestyle factors affecting hypertension. Avoiding smoking, maintaining a balanced diet to prevent unnecessary weight gaining, regular monitoring of blood sugar level for diabetic care, and stress relief exercise can reduce hypertension risk.