Effective incident detection is essential for emergency response and transportation management. Traditional methods relying on stationary technologies are often costly and provide limited coverage, prompting the exploration of crowdsourced data such as Waze. While Waze offers extensive coverage, its data can be unverified and unreliable. This study aims to identify factors affecting the reliability of Waze alerts and develop a predictive model to distinguish true incidents from false alerts using real-time Waze data, thereby improving emergency response times. Real crash data from the New Jersey Department of Transportation (NJDOT) and crowdsourced data from Waze were matched using the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm to differentiate true and false alerts. A binary logit model was constructed to reveal significant predictors such as time categories around peak hours, road type, report ratings, and crash type. Findings indicate that the likelihood of accurate Waze alerts increases during peak hours, on streets, and with higher report ratings and major crashes. Additionally, multiple machine learning-based predictive models were developed and evaluated to forecast in real time whether Waze alerts correspond to actual incidents. Among those models, the Random Forest model achieved the highest overall accuracy (82.5%) and F1-score (82.8%), and an Area Under the Receiver Operating Characteristic Curve (AUC-ROC) of 0.90, demonstrating its robustness and reliability for real-time incident detection. Gradient Boosting, with an AUC-ROC of 0.90 and Area Under the Precision–Recall Curve (AUC-PR) of 0.90, also performed strongly, particularly excelling at predicting true alerts. The analysis further emphasized the importance of key predictors such as time of day, report ratings, and road type. These findings provide actionable insights for enhancing the accuracy of incident detection and improving the reliability of crowdsourced traffic alerts, supporting more effective traffic management and emergency response systems.

On heavily traveled highway corridors, traffic congestion, lane merges, toll facilities, and complex interchanges frequently trigger sudden and aggressive deceleration, commonly referred to as harsh braking (HB). Such maneuvers reflect near-miss driving conditions that may precede crashes. Traditional traffic safety analyses rely primarily on historical crash records, a reactive approach that limits agencies’ ability to identify and address emerging risks in a timely manner. Because HB events are continuously captured by connected-vehicle telematics, they provide an opportunity to evaluate roadway safety risk more proactively. This study investigates the applicability of harsh braking events as a surrogate indicator of crash risk on New Jersey interstate highways. The analysis uses more than 8.5 million connected-vehicle telemetry records from Drivewyze and approximately 45,000 police-reported crashes collected between July and December 2024. HB events were identified using a deceleration threshold of 6 ft/s² (approximately 0.2 g) and spatially matched to one-mile highway segments along with crash data. Spatial analysis shows that both HB events and crashes are highly concentrated along major corridors, including I-95, I-80, I-78, and I-287, with notable clustering near toll plazas and complex interchange areas. Temporal patterns indicate that harsh braking activity increases substantially during late fall, likely reflecting seasonal congestion and adverse weather conditions. To quantify the relationship between HB events and crash frequency, Negative Binomial (NB) and Zero-Inflated Negative Binomial (ZINB) regression models were estimated at the segment level. Results reveal a positive and statistically significant association between HB events and crash counts. In the preferred ZINB model, each additional HB event is associated with approximately a one percent increase in expected crash frequency. While the effect of individual events is small, repeated harsh braking activity corresponds to a meaningful increase in crash risk; for example, an increase of 10 HB events corresponds to an expected crash frequency of about 10% higher. Overall, the findings suggest that connected-vehicle HB data can complement traditional crash records by providing early indications of elevated risk. Incorporating HB monitoring into highway safety programs may support proactive identification of hazardous locations and more timely deployment of targeted countermeasures.

Non-recurrent congestion (NRC), caused by unforeseen events such as crashes, lane closures, and adverse weather, accounts for approximately half of all delays on urban freeways, yet it remains difficult to distinguish from routine congestion at recurrent bottlenecks. This study proposes an NRC detection framework based on a Physics-Informed Neural Network (PINN) that embeds the Lighthill–Whitham–Richards (LWR) conservation law into the learning process to construct a physically consistent baseline of normal traffic states. The traffic flow physics is represented by a two-regime fundamental diagram combining the Greenshields model for free-flow conditions and the Underwood model for congested conditions, and the network is trained by minimizing a composite loss that adaptively balances the data fitting error against the LWR residual. NRC is then detected when the observed density exceeds the PINN-estimated baseline density beyond a tolerance threshold of 150%. The framework was evaluated on a 12 km segment of the Seoul Ring Expressway in Korea using six months of 15 min data collected from seventeen sensor stations. The results show that the proposed model reliably isolates NRC events from recurrent peak-period congestion. From the perspective of symmetry, the framework interprets recurrent traffic as a temporally symmetric background state governed by a conservation law, and non-recurrent congestion as a local breaking of this symmetry, which the physics-constrained residual is designed to expose. The key contribution of this study is a theoretically grounded, label-free anomaly detection approach that couples machine learning with traffic flow theory, offering traffic management centers an automated and interpretable tool for incident detection and response.

Safety Service Patrols (SSPs) are highway assistance programs that provide rapid response to crashes, disabled vehicles, and other incidents, helping to reduce congestion and improve roadway safety. Effective SSP coverage planning is important because timely deployment can facilitate quicker incident clearance and reduce the potential for secondary crashes. However, current SSP operations often rely on fixed patrol routes, which may not fully reflect where crash concentrations vary across time, weather, lighting, and roadway surface conditions. This study develops a data-driven framework to identify candidate priority SSP routes using historical crash records. Major interstate corridors in New Jersey, US, within the SSP coverage area were divided into uniform 1-mile roadway segments, and crashes were matched to segments using spatial proximity, roadway name, road system classification, and travel direction. Segment-level crash frequencies were then evaluated under 1152 scenario combinations defined by time of day, day of week, season, surface condition, weather condition, and lighting condition. Each corridor was represented as a directed graph, and feasible direction-specific routes were ranked based on cumulative crash frequency. The results show that crash concentration is spatially clustered, direction-dependent, and sensitive to operational conditions. Some candidate routes remained stable across scenarios, while others shifted with changes in weather, lighting, surface condition, and temporal period. Validation of the proposed route identification framework using recent crash data produced route overlap scores (ROS) of approximately 74 to 95 percent, indicating that many historically identified high-crash routes remained spatially consistent in the validation period. The framework provides a practical planning tool for SSP route reassessment, shift-level planning, and scenario-based patrol prioritization within existing service areas.