Effective ambulance routing is a top priority when it comes to getting emergency help out to people in need. The traditional go-to approach when it comes to ambulance routing has been to just focus on getting the ambulance to the scene in the shortest time possible, without much thought given to whether the hospital it’s headed for is already overwhelmed with patients or doesn’t have the right resources that need critical care. This study presents a novel, data-driven approach that optimizes ambulance routing by integrating real-time hospital crowdedness forecasting with intelligent routing strategies. Rather than focusing solely on minimizing travel time, this framework takes into account hospital overcrowding, bed availability, and specialized care needs. A hybrid forecasting model is employed to predict hospital crowdedness levels, combining ARIMA (for capturing trends and seasonality), LSTM neural networks (for addressing complex patterns), and Gradient Boosting (for robustness). A Deep Q-Network (DQN) agent, trained over 500 episodes, learns to make dynamic routing decisions, balancing travel distance with hospital readiness. The agent operates within a Markov Decision Process (MDP) to guide ambulances to the hospitals best equipped for specific emergencies. The model will be tested in Montgomery County, PA, using real-world data to evaluate its effectiveness. By optimizing both ambulance routing and hospital preparedness, this innovative approach aims to significantly improve emergency care delivery, ensuring faster, smarter, and more efficient treatment for patients in critical need.

Keywords

Ambulance Routing, Time Series Forecasting, Hospital Crowdedness, Deep Q-Network, ARIMA