An Intelligent Public Safety Threat Prediction Framework Using Deep Q-Learning and Agent-Based Simulation

In modern urban surveillance systems, precision and timely forecasting of safety threats to the people are vital in the prevention of crime and the effective management of emergencies. Combining large-scale urban data with artificial intelligence improves situational awareness and supports smart decision-making. Most current methods rely on fixed predictive models that do not capture, resulting in limited real-time responsiveness and low predictive precision. Such systems are often unable to track dynamic environmental changes and the intricate interactions among urban entities. To address these shortcomings, the paper proposes a new framework, Urban Crime and Emergency Response Management in Real-Time, based on Deep Q-learning and agent-based simulations. The framework combines deep reinforcement learning with an agent-based simulation environment to enable context-aware, adaptive threat prediction. Intelligent agents develop optimal response policies using the Deep Q-Learning algorithm as they continuously interact with a dynamically changing urban environment. The system simulates various emergency scenarios, identifies potential hazards, and recommends timely intervention strategies to enhance operational efficiency. As indicated by the experimental results, the proposed DQL-ABS framework is highly effective in terms of prediction accuracy, reduced response time, and improved coordination between safety management units, compared to traditional static frameworks. These findings indicate the framework's ability to enable proactive, data-driven management of public safety and to enhance smart city infrastructure through adaptive, smart threat mitigation measures.