Real-time Crash Prediction Models Research Articles

Wasserstein Generative Adversarial Network to Address the Imbalanced Data Problem in Real-Time Crash Risk Prediction

Real-time crash risk prediction models aim to identify pre-crash conditions as part of active traffic safety management. However, traditional models which were mainly developed through matched case-control sampling have been criticised due to their biased estimations. In this study, the state-of-art class balancing method known as the Wasserstein Generative Adversarial Network (WGAN) was introduced to address the class imbalance problem in the model development. An extremely imbalanced dataset consisted of 257 crashes and over 10 million non-crash cases from M1 Motorway in United Kingdom for 2017 was then utilized to evaluate the proposed method. The real-time crash prediction model was developed by employing Deep Neural Network (DNN) and Logistic Regression (LR). Crash predictions were performed under different crash to non-crash ratios where synthetic crashes were generated by Wasserstein Generative Adversarial Network (WGAN), Synthetic Minority Over-sampling Technique (SMOTE) and Adaptive Synthetic (ADASYN) sampling respectively. Comparisons were then made with algorithmic-level class balancing methods such as cost-sensitive learning and ensemble methods. Our findings suggest that WGAN clearly outperforms other oversampling methods in terms of handling the extremely imbalanced sample and the DNN model subsequently produces a crash prediction sensitivity of about 70% with a 5% false alarm rate. Based on the findings of this study, proactive traffic management strategies including Variable Speed Limit (VSL) and Dynamic Messing Signs (DMS) could be deployed to reduce the probability of crash occurrence.

Overcoming challenges in crash prediction modeling using discretized duration approach: An investigation of sampling approaches

Until recently, statistical approaches used for real-time crash prediction modeling were limited to case-control design and “sampling of alternatives” approaches. A recent study has developed a duration-based real-time crash prediction model capable of incorporating dynamic (time-varying) covariates within its framework. The modeling approach discretizes the duration between crashes into equal time intervals which can be modeled as alternatives in a multinomial logit framework. The approach, however, requires a reformulation of the original crash dataset to fit its modeling framework which results in considerably large data making model estimation computationally demanding. Additionally, validation of the model in the original study is based on crash data from just one interstate, I-405, assuming homogenous highway segments each 5 miles in length. This study improves upon the original study by investigating sampling techniques that can be applied to the reformulated data to reduce computational load using 2019 crash data from two interstates, I-40 and I-55, in Memphis, Tennessee. Furthermore, discretization of inter-crash duration is undertaken following non-homogenous segmentation of the interstates that is based on highway geometry, terrain, and posted speed limit. To accomplish the study objectives, a relatively small future window of 1 h with 15-minute time intervals is used to discretize the inter-crash duration and obtain the reformulated data. Sampling of crashes for model estimation is then done at the crash, epoch, and segment levels to answer the question of which sampling technique (by crash, epoch, or segment) would result in reasonable accuracy when compared with the complete (100%) data. Results show that 25% of samples drawn at the epoch level can result in a considerable reduction of computational load while providing reasonably consistent estimates.

Real-Time Crash-Risk Optimization at Signalized Intersections

Over the past few decades, numerous adaptive traffic signal control (ATSC) algorithms have been proposed to alleviate traffic congestion and optimize traffic mobility using real-time traffic data, such as data from connected vehicles (CVs). However, most of the existing ATSC algorithms do not consider optimizing traffic safety, likely because of the lack of tools to evaluate safety in real time. In this paper, we propose a novel ATSC algorithm for real-time safety optimization. The algorithm utilizes a traditional Reinforcement Learning approach (i.e., Q-learning) as well as recently developed extreme value theory (EVT) real-time crash prediction models. The algorithm was validated using real-world traffic video data collected from two signalized intersections in British Columbia. The results indicated that, compared with an existing fully actuated signal controller, the developed algorithm can significantly reduce the real-time crash risk by 43% to 45% at the intersection’s approaches even at low CVs market penetration rates.

Transfer learning for spatio-temporal transferability of real-time crash prediction models

Real-time crash prediction is a heavily studied area given their potential applications in proactive traffic safety management in which a plethora of statistical and machine learning (ML) models have been developed to predict traffic crashes in real-time. However, one of the fundamental issues relating to the application of these models is spatio-temporal transferability. The present paper attempts to address this gap of knowledge by combining Generative Adversarial Network (GAN) and transfer learning to examine the transferability of real-time crash prediction models under an extremely imbalanced data setting. Initially, a baseline model was developed using Deep Neural Network (DNN) with crash and microscopic traffic data collected from M1 Motorway in the UK in 2017. The dataset utilised in the baseline model is naturally imbalanced with 257 crash cases and 16,359,163 non-crash cases. To overcome data imbalance issue, Wasserstein GAN (WGAN) was utilised to generate synthetic crash data. Non-crash data were randomly undersampled due to computational limitations. The calibrated model was then applied to predict traffic crashes for five other datasets obtained from M1 (2018), M4 (2017 & 2018 separately) and M6 Motorway (2017 & 2018 separately) by using transfer learning. Model transferability was compared with standalone models and direct transfer from the baseline model. The study revealed that direct transfer is not feasible. However, models become transferable temporally, spatially, and spatio-temporally if transfer learning is applied. The predictability of the transferred models outperformed existing studies by achieving high Area Under Curve (AUC) values ranging between 0.69 and 0.95. The best transferred model can predict nearly 95% crashes with only a 5% false alarm rate by tuning thresholds. Furthermore, the performances of transferred models are on par with or better than the standalone model. The findings of this study proves that transfer learning can improve model transferability under extremely imbalanced settings which helps traffic engineers in developing highly transferable models in future.

A deep reinforcement learning-based intelligent intervention framework for real-time proactive road safety management

We propose a variable speed limit (VSL) system for improving the safety of urban expressways in real time. The system has two main functions: monitoring traffic data and then using the data to assess crash risk through a real-time crash prediction model (RTCPM). When the risk is high, the system triggers VSL control to restore traffic conditions to normal. The study addresses several weaknesses in existing VSL-based real-time safety interventions. Existing models are not widely applicable due to varying detector spacing among different freeways, and even within a study area. Therefore, with the existing detector spacing as an input, a cell transmission model (CTM) is used to simulate traffic states for the desired cell size. A dynamic Bayesian network (DBN) is used for modeling in the RTCPM. The proposed CTM model is then modified to allow VSL control. Whereas existing studies selected various VSL strategies from a predefined list, we employ a deep Q-network, which is a reinforcement learning-based machine learning algorithm, for the VSL control. Two busy segments of the Tokyo Metropolitan Expressway were used as the study area. After several iterations, our proposed real-time system reduced the crash risk by 19%.

A conflict-based approach for real-time road safety analysis: Comparative evaluation with crash-based models

An innovative approach for real-time road safety analysis is presented in this work. Unlike traditional real-time crash prediction models (RTCPMs), in which crash data are used in the training phase, a real-time conflict prediction model (RTConfPM) is proposed. This model can be trained using surrogate measures of safety, and can therefore be applied even in situations in which highly spatial/temporal-accurate crash data are unavailable or unreliable. The application of an RTConfPM consists of using a set of input variables recorded during a given time interval, to predict whether there will be an increased risk of unsafe situations in the following interval.This paper presents an RTConfPM to predict rear-end crashes, using time-to-collision values recorded with radar sensors on multiple motorway cross-sections to define unsafe situations, and traffic conditions recorded on the same sections as input to the model. The RTConfPM is compared to a traditional RTCPM, trained with a dataset of crashes located on the same motorway, and using the same traffic data as input. In both approaches, variable selection is performed with Pearson’s correlation test and random forest; synthetic minority oversampling technique (SMOTE) is used to balance the classes in the training dataset, support vector machine (SVM) is used as classifier, and Monte Carlo cross-validation is adopted for robustness. The two approaches are evaluated considering accuracy, recall, specificity/false alarm rate, and area under the curve (AUC).As shown by the results of this paper, the conflict-based approach appears promising, and is able to predict the occurrence of unsafe situations within 5 min with more than 93% accuracy, recall and specificity, significantly outperforming the RTCPM.

Assessing the crash risks of evacuation: A matched case-control approach applied over data collected during Hurricane Irma

Recent hurricane experiences have created concerns for transportation agencies and policymakers to find better evacuation strategies, especially after Hurricane Irma—which forced about 6.5 million Floridians to evacuate and caused a significant amount of delay due to heavy congestion. A major concern for issuing an evacuation order is that it may involve a high number of crashes in highways. In this study, we present a matched case-control based approach to understand the factors contributing to the increase in the number of crashes during evacuation. We use traffic data for a period of 5 to 10 min just before the crash occurred. For each crash observation, traffic data are collected from two upstream and two downstream detectors of the crash location. We estimate models for three different conditions: regular period, evacuation period, and combining both evacuation and regular period data. Model results show that, if there exist a high volume of traffic at an upstream station and a high variation of speed at a downstream station, the likelihood of crash occurrence increases. Using a panel mixed binary logit model, we also estimate the effect of evacuation itself on crash risk and find that, after controlling for traffic characteristics, during evacuation the chance of a crash is higher than in a regular period. Our findings have implications for evacuation declarations and highlight the need for better traffic management strategies during evacuation. Future studies may develop advanced real-time crash prediction models which would allow us to deploy proactive countermeasures to reduce crash occurrences during evacuation.

A Dynamic Bayesian Network-Based Real-Time Crash Prediction Model for Urban Elevated Expressway

Traffic crash is a complex phenomenon that involves coupling interdependency among multiple influencing factors. Considering that interdependency is critical for predicting crash risk accurately and contributes to revealing the underlying mechanism of crash occurrence as well, the present study attempts to build a Real-Time Crash Prediction Model (RTCPM) for urban elevated expressway accounting for the dynamicity and coupling interdependency among traffic flow characteristics before crash occurrence and identify the most probable risk propagation path and the most significant contributors to crash risk. In this study, Dynamic Bayesian Network (DBN) was the framework of the RTCPM. Random Forest (RF) method was employed to identify the most important variables, which were used to build DBN-based RTCPMs. The PC algorithm combined with expert experience was further applied to investigate the coupling interdependency among traffic flow characteristics in the DBN model. A comparative analysis among the improved DBN-based RTCPM considering the interdependency, the original DBN-based RTCPM without considering the interdependency, and Multilayer Perceptron (MLP) was conducted. Besides, the sensitivity and strength of influences analyses were utilized to identify the most probable risk propagation path and the most significant contributors to crash risk. The results showed that the improved DBN-based RTCPM had better prediction performance than the original DBN-based RTCPM and the MLP based RTCPM. The most probable risk influencing path was identified as follows: speed on current segment (V) (time slice 2)⟶V (time slice 1)⟶speed on upstream segment (U_V) (time slice 1)⟶Traffic Performance Index (TPI) (time slice 1)⟶crash risk on current segment. The most sensitive contributor to crash risk in this path was V (time slice 2), followed by TPI (time slice 1), V (time slice 1), and U_V (time slice 1). These results indicate that the improved DBN-based RTCPM has the potential to predict crashes in real time for urban elevated expressway. Besides, it contributes to revealing the underlying mechanism of crash and formulating the real-time risk control measures.

Crash Prediction on Expressway Incorporating Traffic Flow Continuity Parameters Based on Machine Learning Approach

Real-time crash prediction helps identify and prevent the occurrence of traffic crash. For years, various real-time crash prediction models have been investigated to provide effective information for proactive traffic management. When building real-time crash prediction model, a suitable variable space together with a specific time interval for traffic data aggregation and an appropriate modelling algorithm should be applied. Regarding the intercorrelation problem with variable space, comprehensive real-time crash prediction model considering available traffic data characteristics in applicable circumstances needs to be explored. Taking Xi’an G3001 Expressway as study area, real road traffic and accident data during the period from January 2014 to January 2019 on this expressway are applied for real-time crash prediction. To better capture traffic flow characteristics on expressway and improve the practicality of real-time crash prediction model, two new variables (segment difference coefficient and lane difference coefficient) describing the smoothness and continuity of traffic flow in spatial dimension are developed and incorporated in building the crash prediction model to solve the intercorrelation problem with variable space. Random forest (RF) is then adopted to specify the quantitative relationship between specific variable and crash risk. Real-time crash prediction model based on support vector machine (SVM) using new composed variable space is built. The results show that simplified variable space could contribute to the same classification power in currently used real-time crash prediction models compared with traditional variable space. Moreover, the prediction model based on SVM reaches an accuracy level of 0.9, which performs better than other currently used prediction models.

Multi-Objective reinforcement learning approach for improving safety at intersections with adaptive traffic signal control

Adaptive traffic signal control (ATSC) systems improve traffic efficiency, but their impacts on traffic safety vary among different implementations. To improve the traffic safety pro-actively, this study proposes a safety-oriented ATSC algorithm to optimize traffic efficiency and safety simultaneously. A multi-objective deep reinforcement learning framework is utilized as the backend algorithm. The proposed algorithm was trained and evaluated on a simulated isolated intersection built based on real-world traffic data. A real-time crash prediction model was calibrated to provide the safety measure. The performance of the algorithm was evaluated by the real-world signal timing provided by the local jurisdiction. The results showed that the algorithm improves both traffic efficiency and safety compared with the benchmark. A control policy analysis of the proposed ATSC revealed that the abstracted control rules could help the traditional signal controllers to improve traffic safety, which might be beneficial if the infrastructure is not ready to adopt ATSCs. A hybrid controller is also proposed to provide further traffic safety improvement if necessary. To the best of the authors’ knowledge, the proposed algorithm is the first successful attempt in developing adaptive traffic signal system optimizing traffic safety.

Class-imbalanced crash prediction based on real-time traffic and weather data: A driving simulator study

Objective: Crash occurrence prediction has been of major importance in proactively improving traffic safety and reducing potential inconveniences to road users. Conventional statistical crash prediction models frequently suffer from severe data quality issues and require a significant amount of historical data. On the other hand, even though machine learning (ML) based algorithms have proven to be powerful in predicting future outcomes in different fields of applications, they likely fail to provide satisfactory results unless a tuning parameter approach is conducted. The main objective of this article is to develop real-time crash prediction models that will potentially be employed within traffic management systems.Methods: In this study, two highly optimized data-driven models for crash occurrence prediction have been designed based on the popular machine learning techniques, Support Vector Machine (SVM) and deep neural network Multilayer Perceptron (MLP). To ensure that the proposed algorithms produce robust and stable performance, the optimal scheme for models’ construction has been thoroughly examined and discussed. Additionally, the further boost of models’ performance requires the systemic assessment of crash strongest precursors within the driver-vehicle-environment triptych. Therefore, three categories of features, including driver input responses, vehicle kinematics and weather conditions, were measured during the execution of various driving tasks performed on a desktop driving simulator. Moreover, since crash events typically occur in rare instances tending to be underrepresented in the dataset, an imbalance-aware strategy to overcome the issue was adopted using the Synthetic Minority Oversampling TEchnique (SMOTE).Results: The results show that MLP exhibited the best performing prediction results, most particularly, in clear, overcast and snow conditions, in which MLP recall values were above 94%. Higher F1-score values were achieved in overcast and rain weather by MLP and snow conditions by SVM; whereas over 90% of G-mean levels were obtained under fog and rain conditions for MLP and snow condition for SVM.Conclusion: The findings provide new insights into crash events forecasting and may be used to promote enforcement efforts related to designing crash avoidance/warning systems that enhance the effectiveness of the system’s application based on driver input and vehicle kinematics under various weather conditions.

The importance of flow composition in real-time crash prediction

Previous real-time crash prediction models have scarcely used data disaggregated by vehicle type such as light, heavy and motorcycles. Thus, little effort has been made to quantify the impact of flow composition variables as crash precursors. We analyze the advantages of having access to this data by analyzing two scenarios, namely, with aggregated and disaggregated data. For each case, we build Logistics Regressions and Support Vector Machines models to predict accidents in a major urban expressway in Santiago, Chile. Our results show that having access to disaggregated data by vehicle type increases the prediction power up to 30 % providing, at the same time, much better intuition about the actual traffic conditions that may lead to accidents. These results may be useful when evaluating technology investments and developments in urban freeways.

Real-time crash prediction models: State-of-the-art, design pathways and ubiquitous requirements

Proactive traffic safety management systems can monitor traffic conditions in real-time, identify the formation of unsafe traffic dynamics, and implement suitable interventions to bring unsafe conditions back to normal traffic situations. Recent advancements in artificial intelligence, sensor fusion and algorithms have brought about the introduction of a proactive safety management system closer to reality. The basic prerequisite for developing such a system is to have a reliable crash prediction model that takes real-time traffic data as input and evaluates their association with crash risk. Since the early 21st century, several studies have focused on developing such models. Although the idea has considerably matured over time, the endeavours have been quite discrete and fragmented at best because the fundamental aspects of the overall modelling approach substantially vary. Therefore, a number of transitional challenges have to be identified and subsequently addressed before a ubiquitous proactive safety management system can be formulated, designed and implemented in real-world scenarios. This manuscript conducts a comprehensive review of existing real-time crash prediction models with the aim of illustrating the state-of-the-art and systematically synthesizing the thoughts presented in existing studies in order to facilitate its translation from an idea into a ready to use technology. Towards that journey, it conducts a systematic review by applying various text mining methods and topic modelling. Based on the findings, this paper ascertains the development pathways followed in various studies, formulates the ubiquitous design requirements of such models from existing studies and knowledge of similar systems. Finally, this study evaluates the universality and design compatibility of existing models. This paper is, therefore, expected to serve as a one stop knowledge source for facilitating a faster transition from the idea of real-time crash prediction models to a real-world operational proactive traffic safety management system.

Enhancing the Prediction Performance of Real-Time Crash Prediction Models: A Cell Transmission-Dynamic Bayesian Network Approach

Predicting crash probability in real time is a concept that has inspired studies into complex modeling methods, exploring more sophisticated data collection methods, reducing specification errors by introducing new traffic variables, etc. Most existing real-time crash prediction models (RTCPMs) are based on loop detector data and the model architecture, and their prediction performances are sensitive to the location and layout of detectors with respect to crashes, that is, the spacing of detectors on the road section. Different expressways/freeways have different detector layouts and these vary substantially even within the same expressway/freeway. This limitation is a major obstacle in developing a universal RTCPM. To address this, this paper takes one-minute aggregated loop detector data along with detector layouts as input and employs a cell transmission model (CTM) to interpolate states of traffic flow variables for a pre-defined hypothetical detector layout. Next, several RTCPMs are constructed using Bayesian networks (BNs) and dynamic Bayesian networks (DBNs) for existing and CTM-based detector layouts. It is observed that the CTM can generate traffic flow parameters with a mean percentage error of 12.96%. The results further suggest that the model constructed with the CTM-based uniformly spaced simulated detector data coupled with the DBN method performed better than the model based on the BN and existing non-uniform detector layout.

Predicting Imminent Crash Risk with Simulated Traffic from Distant Sensors

The aim of this research was to investigate the performance of simulated traffic data for real-time crash prediction when loop detector stations are distant from the actual crash location. Nearly all contemporary real-time crash prediction models use traffic data from physical detector stations; however, the distance between a crash location and its nearest detector station can vary considerably from site to site, creating inconsistency in detector data retrieval and subsequent crash prediction. Moreover, large distances between crash locations and detector stations imply that traffic data from these stations may not truly reflect crash-prone conditions. Crash and noncrash events were identified for a freeway section on I-94 EB in Wisconsin. The cell transmission model (CTM), a macroscopic simulation model, was applied in this study to instrument segments with virtual detector stations when physical stations were not available near the crash location. Traffic data produced from the virtual stations were used to develop crash prediction models. A comparison revealed that the predictive accuracy of models developed with virtual station data was comparable to those developed with physical station data. The finding demonstrates that simulated traffic data are a viable option for real-time crash prediction given distant detector stations. The proposed approach can be used in the real-time crash detection system or in a connected vehicle environment with different settings.

The Effectiveness of Traffic Control Measures with Real-time Crash Prediction Model for Urban Expressway Safety

The Effectiveness of Traffic Control Measures with Real-time Crash Prediction Model for Urban Expressway Safety

An optimized real-time crash prediction model on freeway with over-sampling techniques based on support vector machine

An optimized real-time crash prediction model on freeway with over-sampling techniques based on support vector machine

Big Data applications in real-time traffic operation and safety monitoring and improvement on urban expressways

The advent of Big Data era has transformed the outlook of numerous fields in science and engineering. The transportation arena also has great expectations of taking the advantage of Big Data enabled by the popularization of Intelligent Transportation Systems (ITS). In this study, the viability of a proactive real-time traffic monitoring strategy evaluating operation and safety simultaneously was explored. The objective is to improve the system performance of urban expressways by reducing congestion and crash risk. In particular, Microwave Vehicle Detection System (MVDS) deployed on an expressway network in Orlando was utilized to achieve the objectives. The system consisting of 275 detectors covers 75 miles of the expressway network, with average spacing less than 1 mile. Comprehensive traffic flow parameters per lane are continuously archived on one-minute interval basis. The scale of the network, dense deployment of detection system, richness of information and continuous collection turn MVDS as the ideal source of Big Data. It was found that congestion on urban expressways was highly localized and time-specific. As expected, the morning and evening peak hours were the most congested time periods. The results of congestion evaluation encouraged real-time safety analysis to unveil the effects of traffic dynamics on crash occurrence. Data mining (random forest) and Bayesian inference techniques were implemented in real-time crash prediction models. The identified effects, both indirect (peak hour, higher volume and lower speed upstream of crash locations) and direct (higher congestion index downstream to crash locations) congestion indicators confirmed the significant impact of congestion on rear-end crash likelihood. As a response, reliability analysis was introduced to determine the appropriate time to trigger safety warnings according to the congestion intensity. Findings of this paper demonstrate the importance to jointly monitor and improve traffic operation and safety. The Big Data generated by the ITS systems is worth further exploration to bring all their full potential for more proactive traffic management.

Predicting Crashes on Expressway Ramps with Real-Time Traffic and Weather Data

Limited research has been conducted on real-time crash analysis of expressway ramps, although there have been many studies in recent years on estimating real-time crash prediction models for main lines. This study presents Bayesian logistic regression models for single-vehicle (SV) and multivehicle (MV) crashes on expressway ramps by using real-time microwave vehicle detection system data, real-time weather data, and ramp geometric information. The results find that the logarithm of the vehicle count, average speed in a 5-min interval, and visibility are significant factors for the occurrence of SV and MV crashes. The Bayesian logistic regression models show that curved ramps and wet road surfaces would increase the possibility of an SV crash, and off-ramps would result in high risk of MV crashes. The high standard deviation of speed in a 5-min interval would significantly increase MV crash likelihood. Random Forests software was applied in variable importance analysis, and the results revealed that the most important factors influencing crashes on ramps were traffic variables, the second most important factors are weather variables, and the least important but still significant factor was the ramp geometry.

Surrogate safety measure for evaluating rear-end collision risk related to kinematic waves near freeway recurrent bottlenecks

This study presents a surrogate safety measure for evaluating the rear-end collision risk related to kinematic waves near freeway recurrent bottlenecks using aggregated traffic data from ordinary loop detectors. The attributes of kinematic waves that accompany rear-end collisions and the traffic conditions at detector stations spanning the collision locations were examined to develop the rear-end collision risk index (RCRI). Together with RCRI, standard deviations in occupancy were used to develop a logistic regression model for estimating rear-end collision likelihood near freeway recurrent bottlenecks in real-time. The parameters in the logistic regression models were calibrated using collision data gathered from the 6-mile study site between 2006 and 2007. Findings indicated that an additional unit increase in RCRI results in increasing the odds of rear-end collision by 21.1%, a unit increase in standard deviation of upstream occupancy increases the odds by 19.5%, and a unit increase in standard deviation of downstream occupancy increases the odds by 18.7%. The likelihood of rear-end collisions is highest when the traffic approaching from upstream is near capacity state while downstream traffic is highly congested. The paper also reports on the findings from comparing the predicted number of rear-end collisions at the study site using the proposed model with the observed traffic collision data from 2008. The proposed model's true positive rates were higher than those of existing real-time crash prediction models.