Spatiotemporal Demand Forecasting and Resource Optimization
Enhancing Emergency Medical Services through Spatiotemporal Demand Forecasting and Resource Optimization: A Deep Learning Approach
Objective: This study aims to enhance the efficiency of Emergency Medical Services (EMS) by developing a spatiotemporal demand forecasting model using deep learning techniques. The goal is to predict EMS demand more accurately and optimize the allocation of EMS resources across different regions, particularly focusing on areas with healthcare access disparities.
Method and Data: The study uses EMS data from 963,676 cases recorded between January 2016 and December 2021 in Chungcheongnam-do, South Korea. Data on population demographics were sourced from the Korean Statistical Information Service (KOSIS). The methodology employs Spatial-Temporal Graph Convolutional Networks (ST-GCN) to capture the dynamic, regional variations in EMS demand and the Maximum Coverage Location Problem (MCLP) to optimize resource allocation. The model predicts EMS demand hourly and identifies optimal locations for EMS centers based on predicted demand patterns.
Results: The results indicate significant spatial and temporal disparities in EMS demand, particularly in non-metropolitan areas. The ST-GCN model successfully accurately predicted response times, demonstrating a Mean Absolute Error (MAE) of 0.303 minutes. The MCLP optimization suggests that current EMS centers are unevenly distributed, with certain regions showing an over-concentration of resources while others remain underserved. The findings underscore the need for a data-driven reallocation of EMS resources to improve access and reduce response times, particularly for fire-related incidents and in rural areas. This study provides valuable insights for policymakers aiming to optimize EMS resource allocation and improve emergency response in regions with healthcare access disparities.