Anurag Pande

Predicting Freeway Crashes from Loop Detector Data by Matched Case-Control Logistic Regression

Growing concern over traffic safety has led to research into prediction of freeway crashes in an advanced traffic management and information systems environment. A crash likelihood prediction model was developed by using real-time traffic flow variables (measured through a series of underground sensors) potentially associated with crash occurrence. The issues related to real-time application, including range of stations and time slice duration to be examined, were also addressed. The methodology used, matched case-control logistic regression, was adopted from epidemiological studies in which every crash is a case and corresponding noncrashes act as controls. The 5-min average occupancy observed at the upstream station during the 5 to 10 min before the crash, along with the 5-min coefficient of variation in speed at the downstream station during the same time, was found to affect crash occurrence most significantly and hence was used to calculate the corresponding log-odds ratio. A threshold value for this ratio may then be set to determine whether the location must be flagged as a potential crash location. It was shown that by using 1.0 as the threshold for the log-odds ratio, more than 69% crash identification was achieved.

Split Models for Predicting Multivehicle Crashes During High-Speed and Low-Speed Operating Conditions on Freeways

The future of traffic management and highway safety lies in proactive traffic management systems. Crash prediction models that use real-time traffic flow variables measured through a series of loop detectors are the most important component of such systems. A previous crash prediction model was developed with the matched case-control logistic regression technique. Although the model achieved reasonable classification accuracy, it remained open to improvement because of the limited study area, sample size, and transferability issues. Therefore, the previous work had been extended. Multivehicle freeway crashes under high- and low-speed traffic conditions were found to differ in severity and in their mechanism. The distribution of 5-min average speeds obtained immediately before the crash from the loop detector station closest to the crash shows two approximate mound-shaped distributions. This distribution is used as the basis to separate the models for crashes occurring under the two speed conditions. The results show that, as expected, variables that entered in the final models (for crashes under high and low speeds) were not the same. However, they were found to be consistent with the probable mechanisms of crashes under the respective speed conditions. A possible implementation of the separate models with the use of the odds ratios and with the balancing of the threshold between achieving high classification of crash potential and the false alarm situation is presented.

Comprehensive Analysis of the Relationship Between Real-Time Traffic Surveillance Data and Rear-End Crashes on Freeways

Rear-end collisions are the single most frequent type of crash on freeways. Their impact on freeway operation is also most noticeable because almost all of them occur during periods of medium to heavy demand. Preliminary explorations of average traffic speeds before a crash measured at loop detector stations surrounding the crash location showed that rear-end crashes can be placed into two mutually exclusive groups: first, those that occur under extended congestion and, second, those that occur with relatively free-flow conditions prevailing 5 to 10 min before the crash. With loop detector data preceding these two groups of rear-end crashes contrasted with randomly selected noncrash data, it was found that the first group can be attributed to parameters such as the coefficient of variation in speed and average occupancy measurable through loop detectors at stations in the close vicinity of the crash location. For the second group, traffic parameters such as average speed and occupancy at stations downstream of the crash location were significant as were off-line factors such as the time of day and presence of an on-ramp in the downstream direction. It was also observed that traffic conditions belonging to the first segment occurred rarely on the freeway but still made up about half the rear-end crashes. This observation, along with neural network–based classifiers, has been used to propose a strategy for real-time identification of conditions prone to the rear-end crashes. The strategy can potentially identify almost 75% of rear-end crashes, with reasonable false alarms.

Crash Risk Assessment Using Intelligent Transportation Systems Data and Real-Time Intervention Strategies to Improve Safety on Freeways

This article provides a comprehensive overview of the novel idea of real-time traffic safety improvement on freeways. Crash prone conditions on the freeway mainline and ramps were identified using loop detector data, then intelligent transportation systems (ITS) strategies to reduce the crash risk in real-time are proposed. Separate logistic regression models for assessing the risk of crashes occurring under two speed regimes were estimated. The results show that the variables in the two models are consistent with probable mechanisms of crashes under the respective regimes (high-to-moderate and low speed). This study also discusses the analysis of parameters and conditions that affect crash occurrence on freeway ramps by type (on-/off-ramp) and configurations (diamond, loop, etc.) using five-minute traffic flow data obtained from the loop detectors upstream and downstream of ramps to reflect actual traffic conditions prior to the time of crashes. Finally, several traffic management strategies are evaluated for the resulting traffic safety improvement in real-time using PARAMICS microscopic traffic simulation and the measures of crash potential determined through the logistic regression models. The results show that, while variable speed limit strategies reduced the crash potential under moderate-to-high speed conditions, ramp metering strategies were effective in reducing the crash potential during the low-speed conditions.

Assessing Safety on Dutch Freeways with Data from Infrastructure-Based Intelligent Transportation Systems

Most freeway traffic surveillance technologies deployed around the world remain infrastructure based, with underground loop detectors being the most common among them. A proactive application for traffic surveillance data recently explored for some freeways in the United States is the estimation of real-time crash risk. The application involves establishing relationships between historical crashes and archived traffic data collected before those crashes. In these studies, crash occurrence on freeway sections has been related to temporal-spatial variation in speed and high lane occupancy. Critical modeling questions that remain unanswered relate to transferability of such an approach. This study attempts to address the issues of such transfer through analysis of crash data and corresponding loop detector data from five freeways in the Utrecht region of the Netherlands. Traffic surveillance systems for these freeways include more detectors per kilometer than most U.S. freeways. Their real-time data are also already being used for applications of advanced intelligent transportation systems. The analysis procedure proposed here accounts for these distinctions. In addition to these transferability issues, application is introduced of a new data-mining methodology, Random Forests, for identifying variables significantly associated with the binary target variable (crash versus noncrash). It was found that the average and standard deviations of speed and volume are related to real-time crash likelihood. Subjecting these significantly related variables to multilayer perceptron and normal radial basis function neural networks resulted in classifiers that achieved classification accuracy of approximately 61% for crashes and 79% for noncrashes. The promising classification accuracy indicates that these models can be used for reliable assessment of real-time crash risk on Dutch freeways as well.

Estimation of Real-Time Crash Risk: Are All Freeways Created Equal?

In-ground loop detectors have recently been used by many researchers to investigate the links with real-time crash risk and the traffic data. An issue that has been raised, but not explicitly addressed in these studies, is how the results from one freeway might transfer to another. A study was done to examine the relationship between crash risk and real-time traffic variables from a freeway corridor (eastbound I-4 in Orlando, Florida) and then to apply the models to three other freeway corridors (westbound I-4 and northbound and southbound I-95). Traffic data used in the study were collected with loop detectors as well as radar detectors already installed on these freeways. The traffic information was collected for crash as well as random noncrash cases so that a binary classification approach could be adopted. The random forest–based models provided a list of significant variables based on the average reduction in the Gini indices to the overall forest classification. The periods between 5 and 10 min before and between 10 and 15 min before the crash were taken into consideration so that these models could provide the crash risk in advance. Average occupancy of upstream station and average speed and coefficient of variation of volume for downstream stations were found to have a significant effect on crash risk. Application of multilayer perceptron neural network models showed that although the model developed for the I-4 corridor works reasonably well for the westbound I-4 corridor, the performance was not as good for the I-95 sections. This observation indicates that the same model for crash risk identification may work only for corridors with very similar traffic patterns.

Spatiotemporal Variation of Risk Preceding Crashes on Freeways

Research into the application of freeway loop detector data for traffic safety has gained momentum in recent years. The incompleteness of data from loop detectors has been a common problem in both the development and the implementation of models. The effect of individual crash precursors, obtained one at a time from a series of loop detectors, on relative risk of crash occurrence was examined through within-stratum one-covariate logistic regression models. The hazard ratio (resultant change in log odds of observing a crash by changing the covariate by one unit) was used as the measure of risk. The log of coefficient of variation in speed expressed as percentage, standard deviation of volume, and average occupancy expressed as percentage were found to be the most significant individual covariates affecting the odds of crash occurrence at a crash site. It was also observed that these parameters calculated at a 5-min level (as opposed to a 3-min level) are more significantly associated with crash occurrence. Hazard ratios corresponding to these covariates observed at a series of stations during six 5-min slices were plotted as a contour variable. The location and time of measurements of these parameters with respect to the location and time of the crash were used as ordinate and abscissa, respectively, in the contour plots depicting spatiotemporal variation of crash risk. The chart corresponding to the log of coefficient of variation in speed demonstrated the most clear patterns of increasing risk as the time and location of the crash are approached. On the basis of these spatiotemporal patterns, a methodology with which to identify freeway black spots in real time is proposed. This information could be used by traffic management centers to take preventive measures to avoid crashes or to prepare law enforcement and emergency vehicles for the impending situation.

Safety evaluation of multilane arterials in Florida

Resurfacing is one of the more common construction activities on highways. While its effect on riding quality on any type of roadway is obviously positive; its impact on safety as measured in terms of crashes is far from obvious. This study examines the safety effects of the resurfacing projects on multilane arterials with partially limited access. Empirical Bayes method, which is one of the most accepted approaches for conducting before-after evaluations, has been used to assess the safety effects of the resurfacing projects. Safety effects are estimated not only in terms of all crashes but also rear-end as well as severe crashes (crashes involving incapacitating and fatal injuries). The safety performance functions (SPFs) used in this study are negative binomial crash frequency estimation models that use the information on ADT, length of the segments, speed limit and number of lanes. These SPFs are segregated by crash groups (all, rear-end, and severe), length of the segments being evaluated, and land use (urban, suburban, and rural). The results of the analysis show that the resulting changes in safety following resurfacing projects vary widely. Evaluating additional improvements carried out with resurfacing activities showed that all (other than sidewalk improvements for total crashes) of them consistently led to improvements in safety of multilane arterial sections. It leads to the inference that it may be a good idea to take up additional improvements if it is cost effective to do them along with resurfacing. It was also found that the addition of turning lanes (left and/or right) and paving shoulders were two improvements associated with a projects relative performance in terms of reduction in rear-end crashes.

A Novel Approach for Analyzing Severe Crash Patterns on Multilane Highways

This study presents a novel approach for analysis of patterns in severe crashes that occur on mid-block segments of multilane highways with partially limited access. A within stratum matched crash vs. non-crash classification approach is adopted towards that end. Under this approach crashes serve as units of analysis and it does not require aggregation of crash data over arterial segments of arbitrary lengths. Also, the proposed approach does not use information on non-severe crashes and hence is not affected by under-reporting of the minor crashes. Random samples of time, day of week, and location (i.e., milepost) combinations were collected for multilane arterials in the state of Florida and matched with severe crashes from the corresponding corridor to form matched strata consisting of severe crash and non-crash cases. For these cases, geometric design/roadside and traffic characteristics were derived based on the corresponding milepost locations. Four groups of crashes, severe rear-end, lane-change related, pedestrian, and single-vehicle/off-road crashes, on multilane arterials segments were compared separately to the non-crash cases. Severe lane-change related crashes may primarily be attributed to exposure while single-vehicle crashes and pedestrian crashes have no significant relationship with the ADT (Average Daily Traffic). For severe rear-end crashes speed limit, ADT, K-factor, time of day/day of week, median type, pavement condition, and presence of horizontal curvature were significant factors. The proposed approach uses general roadway characteristics as independent variables rather than event-specific information (i.e., crash characteristics such as driver/vehicle details); it has the potential to fit within a safety evaluation framework for arterial segments.

Understanding the Impact of a Recent Hurricane on Mobilization Time During a Subsequent Hurricane

It is not uncommon for a region to be affected by multiple hurricanes in a span of a few weeks. The behavior of the evacuees during a subsequent hurricane in the same season is affected by the damage to the infrastructure and to the vehicles and assets belonging to evacuees, as well as by the psychological impact of the preceding hurricane. One such behavioral aspect that affects traffic-loading rates during a hurricane is the evacuation delay or mobilization time. In this study, “mobilization time for an evacuee” is defined as the difference between the time at which the decision to leave is made and the actual time of departure. This paper proposes a methodology that can be used to understand the factors associated with the mobilization time during a subsequent hurricane while accounting for the effects of the preceding hurricane. The effects of the preceding hurricane were accounted for by modeling mobilization times simultaneously with an ordinal variable representing evacuation participation levels during Hurricane Charley. The data from a survey conducted with the evacuees of Hurricane Frances, which made landfall 3 weeks after Hurricane Charley, were used in this study. The errors for the two simultaneously estimated models were significantly correlated. The results showed that home ownership, the number of individuals in the household, income levels, and the level or the risk of a surge were significant in the model and explained the mobilization times for households. Pet ownership and the number of children in households, known to increase mobilization times during isolated hurricanes, were not found to be significant in the model. The implications of these findings for the demand S-curve are briefly discussed.