Machine Learning Approaches for Region-level Prescription Demand Forecasting

Region-level prescription demand is closely intertwined with the incidence of diseases within a given area. However, conventional forecasting methods primarily rely on historical data, and ignore the spatial correlation in prescription data. In this study, we employ graph structures to capture the interactions among drug demand in different regions. By leveraging two popular graph neural network-based models, our objective is to harness the power of spatial-temporal correlation to enhance the accuracy of predictions. To assess the effectiveness of the graph neural network-based model, we conduct extensive experiments on a comprehensive real world dataset. The results demonstrate that the performance of the graph neural network consistently surpasses that of statistical learning-based methods and traditional deep learning-based methods.