Multi-Context Mining-Based Graph Neural Network for Predicting Emerging Health Risks

Abstract

Patients with similar diseases are able to have similar treatments, care, symptoms, and causes. Based on these relations, it is possible to predict latent risks. Therefore, this study proposes Graph Neural Network-based Multi-Context mining for predicting emerging health risks. The proposed method first, collects and pre-processes chronic disease patients' disease information, behavioral pattern information, and mental health information. After that, it performs context mining. This is a multivariate regression analysis for predicting multiple dependent variables, it extracts a regression model and generates a feature map. Then, the initial graph is created by defining the number of clusters as nodes and constructing edges through correlation. By expanding the graph according to the results of context mining, it is possible to predict that a user has a similar chronic disorder and similar symptoms through users' connection relations. For performance evaluation, the validity of the regression analysis of context mining used in the proposed method, and the suitability of the clustering technique are evaluated.