A multi-stage modeling framework for pedestrian exit choice: Decision-node identification and cross-scenario validation

Abstract

Accurately understanding and predicting pedestrian exit choice during evacuations is a critical scientific issue for enhancing public safety management. Addressing the limitations of existing studies in terms of data realism, incomplete process modeling, and cross-scenario applicability, this paper proposes a multi-stage modeling framework for pedestrian exit choice. Grounded in behavioral science theories, a stage division procedure and identification criteria based on decision nodes are developed. Furthermore, a video-based trajectory extraction and analysis method is introduced to construct large-scale structured pedestrian data, upon which stage-specific discrete choice models are employed to capture exit choice behaviors at different stages. Through cross-scenario case studies of daily routine egress in a metro station and emergency evacuation during an earthquake in a university library, the adaptability and effectiveness of the proposed framework are demonstrated. Results reveal significant scenario and stage dependence in evacuation behavior. Under normal conditions, pedestrian decision-making is guided by convenience, with visibility and distance being dominant factors. In contrast, during emergencies, the decision mode shifts toward efficiency, with flow speed emerging as a key determinant. Moreover, as evacuation progresses into later stages, pedestrians exhibit heightened crowding sensitivity and stronger herding behavior. Differentiated management strategies are proposed accordingly. Bootstrap resampling and agent-based simulations further confirm the robustness and predictive accuracy of the proposed framework. The main contribution of this study lies in establishing a systematic multi-stage evacuation modeling paradigm that links behavioral theory with predictive choice models, quantifies the behavioral shifts across both different scenarios and evacuation stages, and provides a methodological basis for analyzing the entire evacuation process.