Predicting Emergency Medical Service Demand With Bipartite Graph Convolutional Networks
Ruidong Jin, Tianqi Xia, Xin Liu, Tsuyoshi Murata, Kyoung‐Sook Kim
Abstract
Emergency medical service (EMS) plays an essential role in increasing survival rates as it provides first aid to victims of life-threatening emergencies. However, unbalanced EMS supply-demand distribution in the metropolis may cause a shortage of accessible EMS resources and delay the first aid treatment. There is an urgent need to discover the hidden EMS supply-demand relation, predict the incoming EMS demand, and take precautions against unexpected emergencies. This study assumes that EMS demand correlates with population demographic data, regional socioeconomic factors, and hospital conditions. To model these correlated factors, we represent Tokyo's ambulance record data as a hospital-region bipartite graph and propose a bipartite graph convolutional neural network model to predict the EMS demand between hospital-region pairs. Our approach achieves 77.3% - 87.7% accuracy in binary demand label prediction task. It significantly outperforms traditional machine learning algorithms, statistical models, and the latest graph-based methods. Finally, we use a case study to show the significance of EMS demand forecasting, proving that our approach can contribute to public health emergency management by making EMS predictions.