Modeling travelers' safety behaviors in road networks : integrating crash risk models, game theory, and equilibrium analysis

Abstract

Globally, road crashes are a leading cause of fatalities and serious injuries, placing a significant burden on healthcare systems and resulting in substantial economic and social losses. These challenges highlight the critical importance of road network safety planning and policy-making. Road crashes result from a complex interplay of factors, including road design, environmental conditions, vehicle characteristics, and human behavior. Among these, human behavior is the most significant contributor, accounting for 90% of all crashes. Conventional safety planning relies on historical data-based crash prediction models to minimize crash-related costs. While such approaches are helpful in pinpointing factors affecting crash frequency/severity and locating high-risk areas, they fail to incorporate travelers' safety behaviors, preferences, and network-wide interactions, limiting their ability to resolve safety issues proactively. To address these limitations, this thesis proposes integrating safety models with game theory and network equilibrium approaches, enabling the explicit modeling of traveler behaviors and interactions within transportation systems. Such integration can address several key challenges in the field and contribute to developing proactive improvement strategies and policies. First, this thesis examines crash severity prediction modeling techniques, which rely on two main methodological approaches: (a) statistical models and (b) machine learning (ML) models. These approaches aim not only to improve predictive accuracy but also to offer a better understanding of the factors influencing severity. Although ML models often deliver superior or comparable predictive performance relative to statistical models, limited research has explored whether the contributing factors identified by these two approaches are consistent, particularly in the context of vulnerable road users in developing countries. To address these gaps, this thesis utilizes motorcycle crash data from Pakistan. It employs a Shapley value-based cooperative game theory approach to interpret the outputs of ML models in crash severity analysis. The results are compared with those derived from traditional statistical models, revealing that user behavior (e.g., distracted riding) is the most significant factor affecting crash severity. Moreover, the contributing factors identified by ML models align closely with those derived from statistical models, thus enhancing confidence in the interpretability of ML models for transportation safety analysis. These findings strengthen our comprehension of ML applications in safety research and support their use in formulating effective safety policies. Second, this thesis highlights the need for effective screening methods that proactively identify and rank critical links by safety importance, a crucial step for prioritizing interventions and improving overall network safety. Current approaches, such as crash prediction models, identify critical links by analyzing relationships between variables (e.g., roadway characteristics, traffic flow) and safety metrics like crash frequency and severity. However, these methods are limited in evaluating how disruptions on one or more links (e.g., crashes) impact overall network safety due to travelers' rerouting behavior and often ignore network-wide interactions. This thesis addresses these limitations by developing a framework that integrates transportation safety analysis and network equilibrium analysis using a Shapley value-based cooperative game theory approach, where links are treated as players to account for collaborative interactions. A flow-dependent safety evaluation metric, which assesses safety based on varying traffic conditions, is used to calculate the Shapley value. This value represents the average marginal contribution of links to network safety, considering all possible combinations of links known as link coalitions. The results demonstrate that the Shapley value comprehensively captures safety contributions, identifying links connecting multiple origin-destination pairs as having higher safety importance due to greater interactions. This approach provides planners with a valuable tool to identify safety-critical links, enabling improvements in network safety during the planning stage. Third, this thesis emphasizes the importance of enhancing the widely used four-step transportation planning model, which is primarily mobility-based and often overlooks travelers' safety preferences and behaviors. This model involves several key decisions travelers' make, such as whether to travel (trip generation), where to go (trip distribution), which mode of transport to use (modal split) and which route to take (traffic assignment). Car ownership and car type choice are integral to the trip generation and modal split, as they influence the number of trips generated and the modes of transport selected. Given the critical role safety plays in these decisions, as highlighted by numerous empirical studies, it is essential to integrate safety considerations into the transportation planning process. This need is further amplified by the rise of new technologies like autonomous vehicles (AVs), which introduce additional safety and security concerns. To address this, this thesis considers travelers' safety behaviors across various stages of transportation planning using network equilibrium approaches. Specifically, it employs a discrete-choice modeling-based equilibrium analysis framework to examine joint car ownership and car type choice behaviors. It incorporates safety and security risk concerns for both conventional and autonomous vehicles. The findings indicate that safety preferences significantly influence car ownership and car type decisions. Moreover, reductions in travelers' risk perceptions toward AVs lead to significant increases in AV adoption, highlighting the critical role of public trust in transitioning to AV-dominated markets. Additionally, travelers' heterogeneous safety preferences are incorporated into route choice and traffic assignment (TA) models, exploring aspects such as safety-conscious route choice sets, safety reliability requirements, and travelers' heterogeneous value of safety. The results demonstrate that equilibrium flow patterns derived from these models differ from those of conventional mobility-based models, with highly safety-conscious travelers' prioritizing safer routes, even at the cost of longer travel times. These models can serve as a foundation for safety-conscious planning, enabling planners to develop targeted safety improvement policies informed by travelers' safety behaviors, rather than relying solely on historical crash data or past experiences. In summary, this thesis contributes to transportation network safety and planning by integrating safety models, cooperative game theory, and network equilibrium approaches, offering insights into theory and practice. First, it enhances the interpretability of ML-based crash prediction models through the Shapley value, comparing the consistency of crash contributing factors with traditional models and translating these insights into policy recommendations. Second, it develops a transportation network safety analysis framework for identifying the safety critical links, enabling proactive improvements in network safety during the planning stage. Finally, the thesis considers travelers' safety behaviors in car ownership, car type choice, and traffic assignment, shifting from reactive to proactive, behavior-aware strategies. Together, these enhancements provide new insights into network safety, policy development, and safety-conscious transportation planning, offering practical strategies for improving overall network safety.