Real-time crash prediction on express managed lanes of Interstate highway with anomaly detection learning.

To facilitate efficient transportation, I-4 Express is constructed separately from general use lanes in metropolitan area to improve mobility and reduce congestion. As this new infrastructure would undoubtedly change the traffic network, there is a need for more understanding of its potential safety impact. Unfortunately, many advanced real-time crash prediction models encounter an important challenge in their applicability due to their demand for a substantial volume of data for direct modeling. To tackle this challenge, we proposed a simple yet effective approach - anomaly detection learning, which formulates model as an anomaly detection problem, solves it through normality feature recognition, and predicts crashes by identifying deviations from the normal state. The proposed approach demonstrates significant improvement in the Area Under the Curve (AUC), sensitivity, and False Alarm Rate (FAR). When juxtaposed with the prevalent direct classification paradigm, our proposed anomaly detection learning (ADL) consistently outperforms in AUC (with an increase of up to 45%), sensitivity (experiencing up to a 45% increase), and FAR (reducing by up to 0.53). The most performance gain is attained through the combination of Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) in an ensemble, resulting in a 0.78 AUC, 0.79 sensitivity, and a 0.22 false alarm rate. Furthermore, we analyzed model features with a game-theoretic approach illustrating the most correlated features for accurate prediction, revealing the attention of advanced convolution neural networks to occupancy features. This provided crucial insights into improving crash precaution, the findings from which not only benefit private stakeholders but also extend a promising opportunity for governmental intervention on the express lane. This work could promote express lane with more efficient resource allocation, real-time traffic management optimization, and high-risk area prioritization.

Elemental Characterization of Ambient Particulate Matter for a Globally Distributed Monitoring Network: Methodology and Implications.

Global ground-level measurements of elements in ambient particulate matter (PM) can provide valuable information to understand the distribution of dust and trace elements, assess health impacts, and investigate emission sources. We use X-ray fluorescence spectroscopy to characterize the elemental composition of PM samples collected from 27 globally distributed sites in the Surface PARTiculate mAtter Network (SPARTAN) over 2019-2023. Consistent protocols are applied to collect all samples and analyze them at one central laboratory, which facilitates comparison across different sites. Multiple quality assurance measures are performed, including applying reference materials that resemble typical PM samples, acceptance testing, and routine quality control. Method detection limits and uncertainties are estimated. Concentrations of dust and trace element oxides (TEO) are determined from the elemental dataset. In addition to sites in arid regions, a moderately high mean dust concentration (6 µg/m3) in PM2.5 is also found in Dhaka (Bangladesh) along with a high average TEO level (6 µg/m3). High carcinogenic risk (>1 cancer case per 100000 adults) from airborne arsenic is observed in Dhaka (Bangladesh), Kanpur (India), and Hanoi (Vietnam). Industries of informal lead-acid battery and e-waste recycling as well as coal-fired brick kilns likely contribute to the elevated trace element concentrations found in Dhaka.

Exploring the influence of drivers' visual surroundings on speeding behavior.

Despite awareness campaigns and legal consequences, speeding is a significant cause of road accidents and fatalities globally. To combat this issue, understanding the impact of a driver's visual surroundings is crucial in designing roadways that discourage speeding. This study investigates the influence of visual surroundings on drivers in 15 US cities using 3,407,253 driver view images from Lytx, covering 4,264 miles of roadways. By segmenting and analyzing these images along with vehicle-related variables, the study examines factors affecting speeding behavior. After filtering the images, to ensure an accurate representation of the driver's view, 1,340,035 driver view images were used for analysis. Statistical models, including hurdle beta and bivariate probit models with random driver effects as well as Machine Learning's eXtreme Gradient Boosting (XGBoost), were employed to estimate speeding behavior. The results indicate that factors within the driver's visual environment, weather conditions, and driver heterogeneity significantly impact speeding. Speeding behavior also varies across geographic locations, even within the same city, suggesting a connection between local context and speeding. The study highlights the importance of the driver's environment, showing that more open spaces encourage speeding, while areas with trees and buildings are associated with reduced speeding. Notably, this research differs from previous studies by utilizing real-time data from dash cameras, providing a dynamic and accurate representation of the driver's visual surroundings. This approach enhances the reliability of the findings and empowers transportation engineers and planners to make informed decisions when designing roadways and implementing interventions to address effectively excessive speeding. In addition to examining speeding behavior, the study also analyzes time-headway, a key factor affecting safety and risky driver behavior, to explore its relationship with speeding. The findings offer valuable insights into the factors influencing speeding and the driver's visual environment. These insights can inform efforts to create environments that discourage speeding (and close car following) and ultimately reduce severe accidents caused by excessive speed (and tailgating).

Understanding the impact of vehicle dynamics, geometric and non-geometric roadway attributes on surrogate safety measure using connected vehicle data.

Traditional safety research mostly relies on accident data to analyze the precedents to a crash. Alternatively, surrogate safety measures have the potential to proactively evaluate safety events. The era of connected vehicles and smart sensing has brought about tremendous innovations in safety research. GPS data from such vehicles form a useful case of big data analytics where surrogate safety measures have largely been unexplored. In this paper, we propose time to collision estimation from connected vehicle GPS data. The vehicle dynamics such as speed, acceleration, yaw rate, etc. are then coupled with geometric and non-geometric roadway attributes to understand the contributing factors for a traffic conflict. The dataset contains 2,568,421 GPS points from 14,753 unique journeys. 1:4 ratio of conflict to non-conflict events was used to select 15,258 samples with 28 independent vehicle dynamics, geometric, and non-geometric variables. Binary logit model was used to investigate the relationship of these variables with conflicts. Model results showed that out of 28 independent variables, 6 independent variables and 7 interaction variables were found significant. The results showed some interesting and unique relations of these variables with conflicts. Based on these significant variables, k-means clustering was performed to understand the threshold for the significant values for which the number of conflicts is significantly increased. Results from k-means clustering and two sample binomial proportion t-tests revealed that when absolute acceleration crossed 0.8 m/s2, conflict probability increased by 8 percentage points.​ Moreover, when the yaw rate crossed 8 degrees/s, the conflict probability doubled. Besides, vehicles traveling at more than 140% of the recommended speed limit increased conflict probability by 7 percentage points.

Effect of signal timing on vehicles' near misses at intersections.

Driving characteristics often vary between the different states of the signal. During red and yellow phase, drivers tend to speed up and reduce the following distance which in turn increases the possibility of rear end crashes. Intersection safety, therefore, relies on the correct modelling of signal phasing and timing parameters, and how drivers respond to its changes. This paper aims to identify the relationship between surrogate safety measures and signal phasing. Unmanned aerial vehicle (UAV) video data has been used to study a major intersection. Post encroachment time (PET) between vehicles was calculated from the video data as well as speed, heading and relevant signal timing parameters such as all red time, red clearance time, yellow time, etc. Random parameter ordered logit model was used to model the relationship between PET and signal timing parameters. Overall, the results showed that yellow time and red clearance time is positively related to PETs. The model was also able to identify certain signal phases that could be a potential safety hazard and would need to be retimed by considering the PETs. The odds ratios from the models also indicate that increasing the mean yellow and red clearance times by one second can improve the PET levels by 10% and 3%, respectively.

Factors affecting injury severity at pedestrian crossing locations with Rectangular RAPID Flashing Beacons (RRFB) using XGBoost and random parameters discrete outcome models.

This paper evaluates the effectiveness of Rectangular Rapid Flashing Beacons (RRFB) on crash severity. The study used and compared XGBoost and Random Parameters Discrete Outcome Models (RPDOM) respectively. The dataset comprises of 312 pedestrian crossing locations, among which 154 treatment locations were provided with the Rectangular Rapid Flashing Beacons (RRFB) and 158 control locations without RRFB. These control locations have similar roadway, traffic, and land use characteristics of that of the treatment locations but are not treated with RRFB or other pedestrian crossing countermeasures. This study shows the impact of RRFB and other factors on severity of nighttime, pedestrian, total and rear-end crashes. Crash severity data was compiled from driver, vehicle, and event level data of each crash. Due to availability of larger number of observations for total (35,553), rear-end (15,675) and nighttime crashes (8,144) XGBoost was used, and due to less observations for pedestrian crashes (369), it was modeled using RPDOM. The results showed positive impact of RRFB for the reduction of nighttime crashes. It was noted that RRFB reduces the K and A nighttime crashes according to the SHAP values from the XGBoost model but does not have the desired significance for rear end and overall total crashes in the study area. From the RPDOM, it was seen that RRFB showed statistically significant reduction in injury severity of pedestrian crashes and nighttime crashes. To compare the two models, nighttime crashes were modeled using both the techniques, the prediction accuracy of XGBoost Model was 97% which was much greater than that of the RPDOM at 73.8% prediction accuracy. Thus, both XGBoost and the RPDOM model for showed positive impact of installing RRFB in reducing the severity of nighttime crashes.

Crash data augmentation using variational autoencoder.

In this paper, we present a data augmentation technique to reproduce crash data. The dataset comprising crash and non-crash events are extremely imbalanced. For instance, the dataset used in this paper consists of only 625 crash events for over 6.5 million non-crash events. Thus, learning algorithms tend to perform poorly on these datasets. We have used variational autoencoder to encode all the events into a latent space. After training, the model could successfully separate crash and non-crash events. To generate data, we sampled from the latent space containing crash data. The generated data was compared with the real data from different statistical aspects. t-Test, Levene-test and Kolmogrove Smirnov test showed that the generated data was statistically similar to the real data. It was also compared to some of the minority oversampling techniques like SMOTE and ADASYN as well as the GAN framework for generating data. Crash prediction models based on Logistic Regression (LR), Support Vector Machine (SVM) and Artificial Neural Network (ANN) were used to compare the generated data from the different oversampling techniques. Overall, variational autoencoder (VAE) showed excellent results compared to the other data augmentation methods. Specificity is improved by 8% and 4% for VAE-LR and VAE-SVM respectively when compared to SMOTE while the sensitivity is improved by 6% and 5% when compared to ADASYN. Moreover, VAE generated data also helps to overcome the overfitting problem in SMOTE and ADASYN since there is flexibility in choosing the decision boundary.