IRIS-AI: An AI-Based Resource Recommender for Emergency Decision Support Systems of Fire Departments

In environments as dynamic and high-stakes as emergency scenarios, Emergency Decision Support Systems (EDSS) have been used to increase the effectiveness of emergency services response by compiling the large volume of information about the incident state in a single system that assists first responders in making timely and accurate decisions. Nonetheless, Incident Commanders can further benefit from machine-generated recommendations based on forecasting of future evolution of incidents and the adequate response to that evolution. Previous works on providing such recommendations have targeted large-scale disasters and public emergencies, limiting the generalization of these approaches and their wide adoption. In this paper, we introduce IRIS-AI, a novel resource recommendation system that can be applied across emergency incident types faced by fire departments. IRIS-AI decouples an incident type from resource characteristics and leverages state-of-the-art AI models to forecast an emergency incident’s imminent specific needs according to its continuously-changing state, thus achieving cross-scenario applicability. We evaluate IRIS-AI on a diverse dataset coming from a real-world EDSS with different types of emergency incidents. The experiments demonstrate the accuracy of our AI models with a maximum $$R_2$$ score of 0.89, as well as the usability of our approach in accompanying Incident Commanders in the resource allocation process by combining it with a Mixed-Integer Linear Program that was able to recommend optimal resources in 82% of test incidents.