Multi-context path-aware graph convolutional neural network learning for health risk prediction

Existing works on graph convolutional neural network learning methods for health risk prediction indicate that there are limitations due to a lack of multimodal fusion using text, Electronic Health Records (EHR), lifestyle data and Internet of Things (IoT) data. Other limitations include the use of heterogeneous graph neural networks, absence of path-aware graph reasoning, and weak patient-similarity modeling.  Moreover, the class imbalance was not adequately addressed. Few studies apply focal loss optimization for mental-health risk prediction, and there are limited multiclass risk prediction methods with classes such as Low, Moderate, High, and Critical. To address these limitations, we propose a multi-context path-aware graph convolutional neural network learning approach that utilizes BERT and GCNN for health risk prediction. This research enhances the performance of health risk prediction specifically for patients with mental health issues.