A generative AI–driven intelligent rules framework for traffic accident detection and analysis

• This paper introduces A Generative AI–Driven Intelligent Rules Framework with Time-to-Collision (TTC) metrics to improve the accuracy of collision prediction in cut-in scenarios, and the development of a comprehensive collision avoidance strategy. • The goals of this research are to develop a reliable prediction system for detecting potential collisions during cut-in scenarios, propose suitable evasive maneuvers such as deceleration or lane changes, and ultimately enhance the safety of automated vehicles (AVs) based generative test to improve their collision avoidance capabilities in complex traffic environments. • The novelty of this research lies in combining Rules −Based Approach with behavioral insights and dynamic TTC computation to create a comprehensive, adaptive collision avoidance framework that addresses limitations in static, rule-based systems. Most road accidents are primarily caused by human factors and driver errors. Young and inexperienced drivers are disproportionately involved in crashes, with inattentive or aggressive driving behaviors further increasing accident risk, especially during lane-changing maneuvers that unpredictably reduce the available gap between vehicles. Such behaviors also challenge the effectiveness of Advanced Driver Assistance Systems (ADAS), which rely on stable and predictable interactions to maintain safety. Recent advances in Generative Artificial Intelligence (Generative AI) have introduced new possibilities for intelligent simulation and analysis of driving scenarios. This paper proposes a Generative AI–driven intelligent rules framework for modeling, simulating, and interpreting traffic accident dynamics. The framework combines generative modeling, capable of synthesizing realistic and diverse accident scenarios with rule-based reasoning that incorporates domain knowledge and driver behavior models. Through simulation-based scenario generation and intelligent rule inference, the system enables both quantitative and qualitative analysis of accident causality and risk propagation. Experimental validation using real-world driving data demonstrates the framework’s capability to replicate rare but safety–critical events, uncover hidden causal links between vehicle dynamics and collision risk, and support adaptive, data-driven safety decisions. Furthermore, the study emphasizes the Time-to-Collision (TTC) metric as a key safety indicator for evaluating driver attentiveness, reaction behavior, and system response under varying traffic conditions. To address the increasing complexity of modern road environments, this work advocates for the integration of hybrid intelligence merging human cognitive modeling with generative AI to enhance situational awareness and decision-making across diverse driving contexts. Special attention is given to the challenges of mixed traffic management, where variations in cultural norms, behavioral tendencies, and external stressors necessitate adaptive safety strategies grounded in behavioral and contextual awareness.