Competing risks mixture model for traffic incident duration prediction

Unraveling the determinants of traffic incident duration: A causal investigation using the framework of causal forests with debiased machine learning

Predicting the duration of traffic incidents is challenging due to their stochastic nature. Accurate predictions can greatly benefit end-users by informing their route choices and safety warnings, while helping traffic operation managers more effectively manage non-recurrent traffic congestion and enhance road safety. This study conducts a comprehensive causal analysis of traffic incident duration using a data collected over a long time and including different types of roads across the city of Tianjin, China. Employing the innovative framework of causal forests with biased machine learning (CF-DML) techniques, this study advances beyond traditional methods by focusing on interpreting the causal relationships between various factors and incident duration, emphasizing the role of heterogeneity among these factors. The CF-DML framework enables the assessment of the average treatment effects (ATEs) of various factors on incident duration. Notably, the significant influence of road type and suburban setting on treatment effects is underscored, which is generally consistent with the results obtained through classical methods. Second, to look more closely at the important factors such as road and collision types, a conditional average treatment effects (CATE) analysis is conducted, explaining heterogeneity through a causal heterogeneity tree. Third, based on insights from causal analysis, policies related to lane configurations are explored, emphasizing the necessity of considering causal effects in traffic management decisions. The CF-DML framework enhances our understanding of traffic incident dynamics, contributing to improved road safety and traffic flow in diverse urban environments.

Study on ring-road incident duration based on latent class accelerated hazard model

Accurately estimating the duration of freeway incidents can enhance emergency management practices and reduce the likelihood of secondary incidents. To investigate the mechanisms through which key factors influence incident duration, this study sorted out the characteristics and variables of the incident duration on a special freeway in Zhejiang Province, that is, the ring road, and developed a latent class accelerated hazard model. Heterogeneity was incorporated into the model. Three distributions (Weibull, Log-normal, and Log-logistic) were compared, and the Log-logistic distribution exhibited superior performance. The analysis revealed two distinct latent classes: Latent Class 1 and Class 2, had class membership probability of 0.53 and 0.47, respectively, with a total of 11 variables being statistically significant at the 0.05 significance level. It is worth noting that, some neglected explanatory variables are discussed in depth in this study. For example, the mechanism of which specific lane is closed has an impact on the incident duration, rather than a general discussion of the number of lane closures. Furthermore, the way in which the driver involved in the incident reports to the police has a significant impact on the duration of incidents. Notably, potential heterogeneity and its influencing mechanism are captured in the model. Additionally, by predicting class membership using posterior probabilities, it was determined that most data points were more likely to belong to Class 1, and the incident duration primarily ranged between 0 and 60 minutes. These findings are helpful to reduce the duration of incidents on ring-roads and freeways in China, and provide theoretical support for the formulation of freeway incident management and treatment policies.

Traffic accident duration prediction using multi-mode data and ensemble deep learning

Predicting the duration of traffic accidents is a critical component of traffic management and emergency response on expressways. Traffic accident information is inherently multi-mode data in terms of data types. However, most existing studies focus on single-mode data, and the influence of multi-mode data on the prediction performances of models has been the subject of only very limited quantitative analysis. The present work addresses these issues by proposing a heterogeneous deep learning architecture employing multi-modal features to improve the accuracy of predictions for traffic accident durations on expressways. Firstly, six unique data modes are obtained based on the structured data and the text data. Secondly, a hybrid deep learning approach is applied to build classification models with reduced prediction error. Finally, a rigorous analysis of the influence for multi-mode data on the accident duration prediction performances is conducted using a variety of deep learning models. The proposed method is evaluated using survey data collected from an expressway monitoring system in Shaanxi Province, China. The experimental results show that Word2Vec-BiGRU-CNN is a suitable and better model using text features for traffic accident duration prediction, as the F1-score is 0.3648. This study confirms that the newly established structured features extracted from text data substantially enhance the prediction effects of deep learning algorithms. However, these new features were a detriment to the prediction effects of conventional machine learning algorithms. Accordingly, these results demonstrate that the processing and extraction of text features is a complex issue in the field of traffic accident duration prediction.

Prediction of Duration of Traffic Incidents by Hybrid Deep Learning Based on Multi-Source Incomplete Data

Traffic accidents causing nonrecurrent congestion and road traffic injuries seriously affect public safety. It is helpful for traffic operation and management to predict the duration of traffic incidents. Most of the previous studies have been in a certain area with a single data source. This paper proposes a hybrid deep learning model based on multi-source incomplete data to predict the duration of countrywide traffic incidents in the U.S. The text data from the natural language description in the model were parsed by the latent Dirichlet allocation (LDA) topic model and input into the bidirectional long short-term memory (Bi-LSTM) and long short-term memory (LSTM) hybrid network together with sensor data for training. Compared with the four benchmark models and three state-of-the-art algorithms, the RMSE and MAE of the proposed method were the lowest. At the same time, the proposed model performed best for durations between 20 and 70 min. Finally, the data acquisition was defined as three phases, and a phased sequential prediction model was proposed under the condition of incomplete data. The results show that the model performance was better with the update of variables.

Comparing and Contrasting the Impacts of Macro-Level Factors on Crash Duration and Frequency

Road traffic crashes cause social, economic, physical and emotional losses. They also reduce operating speed and road capacity and increase delays, unreliability, and productivity losses. Previous crash duration research has concentrated on individual crashes, with the contributing elements extracted directly from the incident description and records. As a result, the explanatory variables were more regional, and the effects of broader macro-level factors were not investigated. This is in contrast to crash frequency studies, which normally collect explanatory factors at a macro-level. This study explores the impact of various factors and the consistency of their effects on vehicle crash duration and frequency at a macro-level. Along with the demographic, vehicle utilisation, environmental, and responder variables, street network features such as connectedness, density, and hierarchy were added as covariates. The dataset contains over 95,000 vehicle crash records over 4.5 years in Greater Sydney, Australia. Following a dimension reduction of independent variables, a hazard-based model was estimated for crash duration, and a Negative Binomial model was estimated for frequency. Unobserved heterogeneity was accounted for by latent class models for both duration and frequency. Income, driver experience and exposure are considered to have both positive and negative impacts on duration. Crash duration is shorter in regions with a dense road network, but crash frequency is higher. Highly connected networks, on the other hand, are associated with longer length but lower frequency.

A comparative analysis of freeway crash incident clearance time using random parameter and latent class hazard-based duration model

The effects of freeway incident clearance times on the flow of traffic have recently increased interests in understanding what factors influence incident durations. This has particularly become topical due to the financial and economic implications of traffic gridlocks caused by freeway incidents on industries and personal mobility. This paper presents two advanced econometric modeling methods, random parameters duration modeling and latent class duration modeling in understanding the factors that impact freeway incident clearance times in the State of Alabama. These two modeling approaches were further compared to identify which of them provides the best fit for the data with respect to accounting for unobserved heterogeneity. A total of 2206 freeway crash incident data from January 1 to December 31, 2018 were examined in developing the models. The study was based on a unique dataset that involved merging and matching Traffic Incident Management response data from the Alabama Department of Transportation (ALDOT) Traffic Management Center (TMC), freeway crash data from the Center for Advanced Public Safety (CAPS) at the University of Alabama, Alabama Service and Assistance Patrol (ASAP) data from ALDOT and traffic volume from ALDOT's Highway Performance Management System (HPMS). The model estimation results reveal that a total of nineteen variables were found statistically significant with five random variables (on-road, nighttime, rain, AADT, and ASAP existing coverage area) and fourteen fixed effects variables for the random parameters model. For latent class model, a total of eighteen variables were observed statistically significant within two distinct latent classes (Latent Class 1 with class membership probability of 0.23 and Latent Class 2 with class membership probability of 0.77) at a 0.05 significance level. A comparison of the two models reveals that the latent class model provides the better fit for the incident duration data. The findings of this study are expected to contribute to the body of knowledge on incident duration by employing two advanced econometric modeling methods and to inform statewide efforts in significantly reducing the duration of freeway incident clearance time. Moreover, this is to ensure that policy decisions that may arise from the findings of the study are sound and based on data-driven evidence.

Incorporating multiple congestion levels into spatiotemporal analysis for the impact of a traffic incident

Traffic incidents occurring on the road interrupt the smooth mobility of traffic flow and lead to traffic congestion. Although there has been a proliferation of studies that attempt to estimate the spatiotemporal impact of a traffic incident, most, if not all, of them focus exclusively on the differentiation of bi-level traffic status. In this research, we propose to incorporate multiple congestion levels in the spatiotemporal analysis for the impact of a traffic incident, which is new to the literature. The input to our model includes the historical speed on a given road and the occurrence time and location of the incident. The model then outputs the spatiotemporal impact region with multiple congestion levels. We first use a discriminant indicator to initially indicate the traffic status according to the travelling speed of probe vehicles. We then develop an integer programming model with a set of novel constraints to estimate the spatiotemporal region impacted by the incident. Unlike existing studies that only distinguish between uncongested and congested status, our model allows us to determine the impact region with diverse congestion levels. We validate our model using both simulation and real data. Results demonstrate that our approach can not only ensure the consistency of the propagation of shockwaves even when multiple congestion levels are incorporated, but also produce more accurate estimation of the delay caused by the incident when compared to the current state-of-the-art approach.

Prediction of Loss of Position during Dynamic Positioning Drilling Operations Using Binary Logistic Regression Modeling

The prediction of loss of position in the offshore industry would allow optimization of dynamic positioning drilling operations, reducing the number and severity of potential accidents. In this paper, the probability of an excursion is determined by developing binary logistic regression models based on a database of 42 incidents which took place between 2011 and 2015. For each case, variables describing the configuration of the dynamic positioning system, weather conditions, and water depth are considered. We demonstrate that loss of position is significantly more likely to occur when there is a higher usage of generators, and the drilling takes place in shallower waters along with adverse weather conditions; this model has very good results when applied to the sample. The same method is then applied for obtaining a binary regression model for incidents not attributable to human error, showing that it is a function of the percentage of generators in use, wind force, and wave height. Applying these results to the risk management of drilling operations may help focus our attention on the factors that most strongly affect loss of position, thereby improving safety during these operations.

Real-time traffic accidents post-impact prediction: Based on crowdsourcing data

Traffic accident management is a critical issue for advanced intelligent traffic management. The increasingly abundant crowdsourcing data and floating car data provide new support for improving traffic accident management. This paper investigates the methods to predict the complicated behavior of traffic flow evolution after traffic accidents using crowdsourcing data. Based on the available data source, the traffic condition is divided into four levels by congestion delay index: severely congested, congested, slow moving and uncongested. Four types of accidents are consequently defined based on the occurrence of each level. A hierarchical scheme is designed for identifying the most congested level and sequentially predicting duration of each level. The proposed model is validated using traffic accident data in 2017 from an anonymous source in Beijing, China by embedding three machine learning algorithms, random forest (RF), support vector machine (SVM) and neural network (NN), in the scheme. The results show NN outperforms the other two models when the assessment is conducted in absolute differences. Meanwhile, RF has a slightly better performance than SVM, especially when predicting the short-period congestion of severely congested level at the first time. By continuously updating the traffic condition information, significant improvement in accuracy can be acquired regardless of the exact model used. This study shows that emerging crowdsourcing data can be used in a real-time analysis of traffic accidents and the proposed model is effective to analyze such data.

Effects of the penalty mechanism against traffic violations in China: A joint frailty model of recurrent violations and a terminal accident

A penalty mechanism is usually considered as a powerful means to reduce the probability of traffic violations and accidents by encouraging drivers to comply with traffic regulations. Penalty point and fine strategies are often used in parallel. Different degrees of penalty points and/or fines are imposed according to the specific violation behavior of drivers. However, the question of whether each penalty produces positive effects in maintaining a driver's compliance with traffic regulations and promoting the driver's traffic safety is still unanswered. This study focuses on quantifying the effects of penalty point and fine strategies on violation recurrences and accident occurrences of drivers. A frailty survival analysis method is conducted to jointly model the occurrence of violation and accident events of each individual. The frailty term in the model is leveraged to address the unobserved heterogeneity among drivers. Personal characteristics and penalty status are also incorporated as covariates in the model. Actual violation and accident data from a province in China are utilized to calibrate the model. The results show that penalty point strategy exhibits deterrent and binding effects; however, penalty fine strategy does not show the expected effects. The number of years of driving is also a significant factor that influences violation recurrence and accident occurrence. The present study provides insightful information for improving violation penalty mechanisms.

Revealing Recurrent Urban Congestion Evolution Patterns with Taxi Trajectories

Urban congestion can be classified into two types: Recurrent Congestion (RC) and Non-Recurrent Congestion (NRC). RC is more regular than NRC, having fixed and long-standing patterns. Mining urban recurrent congestion evolution patterns can assist with congestion cause analysis and the creation of alleviating strategies. Most existing methods for analyzing urban congestion patterns are based on traditional traffic detector data, which is inflexible and expensive. Additionally, prior research primarily focused on the microscopic model, which simulated congestion propagation based on theoretical models and hypothetical networks. As such, most previous models and methods are difficult to apply to real case scenarios. Therefore, we investigated recurrent congestion patterns by mining historical taxi trajectory data that were collected in Harbin, China. A three-step method is proposed to reveal urban recurrent congestion evolution patterns. Firstly, a grid-based congestion detection method is presented by calculating the change in taxi global positioning system (GPS) trajectory patterns. Secondly, a customized cluster algorithm is applied to measure the recurrent congestion area. Finally, a series of indicators are proposed to reflect RC evolution patterns. A case study was competed in the Harbin urban area to evaluate the main methods. Finally, RC cause analysis and alleviating strategy are discussed. The results study are expected to provide a better understanding of urban RC evolution patterns.

Exploring the influential factors in incident clearance time: Disentangling causation from self-selection bias

Understanding the relationships between influential factors and incident clearance time is crucial to make effective countermeasures for incident management agencies. Although there have been a certain number of achievements on incident clearance time modeling, limited effort is made to investigate the relative role of incident response time and its self-selection in influencing the clearance time. To fill this gap, this study uses the endogenous switching model to explore the influential factors in incident clearance time, and aims to disentangle causation from self-selection bias caused by response process. Under the joint two-stage model framework, the binary probit model and switching regression model are formulated for both incident response time and clearance time, respectively. Based on the freeway incident data collected in Washington State, full information maximum likelihood (FIML) method is utilized to estimate the endogenous switching model parameters. Significant factors affecting incident response time and clearance time can be identified, including incident, temporal, geographical, environmental, traffic and operational attributes. The estimate results reveal the influential effects of incident, temporal, geographical, environmental, traffic and operational factors on incident response time and clearance time. In addition, the causality of incident response time itself and its self-selection correction on incident clearance time are found to be indispensable. These findings suggest that the causal effect of response time on incident clearance time will be overestimated if the self-selection bias is not considered.