Multi-vehicle safety functions for freeway weaving segments using lane-level traffic data

The negative Binomial-Lindley model with Time-Dependent Parameters: Accounting for temporal variations and excess zero observations in crash data

Crash counts are non-negative integer events often analyzed using crash frequency models such as the negative binomial (NB) distribution. However, due to their random and infrequent nature, crash data usually exhibit unique characteristics, such as excess zero observations that the NB distribution cannot adequately model. The negative binomial-Lindley (NBL) and random parameters negative binomial-Lindley (RPNBL) models have been proposed to address this limitation. Despite addressing the issues of excess zero observations, these models may not fully account for unobserved heterogeneity resulting from temporal variations in crash data. In addition, many variables, such as traffic volume, speed, and weather, change with time. Therefore, the analyst often requires disaggregated data to account for their variations. For example, it is recommended to use monthly crash datasets to better account for temporally varying weather variables compared to yearly crash data. Using temporally disaggregated data not only adds the complexity of the temporal variations issue in data but also compounds the issue of excess zero observations. To address these issues, this paper introduces a new variant of the NBL model with coefficients and Lindley parameters that vary by time. The derivations and characteristics of the model are discussed. Then, the model is illustrated using a simulation study. Subsequently, the model is applied to two empirical crash datasets collected on rural principal and minor arterial roads in Texas. These datasets include several time-dependent variables such as monthly traffic volume, standard deviation of speed, and precipitation and exhibit unique characteristics such as excess zero observations. The results of several goodness-of-fit (GOF) measures indicate that using the NBL model with time-dependent parameters enhances the model fit compared to the NB, NBL, and the NB model with time-dependent parameters. Findings derived from crash data collected from both rural minor and principal arterial roads in Texas suggest that the variables denoting the median presence and wider shoulder width are associated with a potential decrease in crash occurrences. Moreover, higher variations in speed and wider road surfaces are linked to a potential increase in crash frequency. Similarly, a higher monthly average daily traffic (Monthly ADT) positively correlates with crash frequency. We also found that it is important to account for temporal variations using time-dependent parameters.

A study on diversion behavior in weaving segments: Individualized traffic conflict prediction and causal mechanism analysis

Lane change behavior disrupts traffic flow and increases the potential for traffic conflicts, especially on expressway weaving segments. Focusing on the diversion process, this study incorporating individual driving patterns into conflict prediction and causation analysis can help develop individualized intervention measures to avoid risky diversion behaviors. First, to minimize measurement errors, this study introduces a lane line reconstruction method. Second, several unsupervised clustering methods, including k-means, agglomerative clustering, gaussian mixture, and spectral clustering, are applied to explore diversion patterns. Moreover, machine learning methods, including Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), Attention-based LSTM, eXtreme Gradient Boosting (XGB), Support Vector Machine (SVM), and Multilayer Perceptron (MLP), are employed for real-time traffic conflict prediction. Finally, mixed logit models are developed using pre-conflict condition data to investigate the causal mechanisms of traffic conflicts. The results indicate that the K-means algorithm with four clusters exhibits the highest Calinski-Harabasz and Silhouette scores and the lowest Davies-Bouldin scores. With superior classification accuracy and generalization ability, the LSTM is used to develop the personalized traffic conflict prediction model. Sensitivity analysis indicates that incorporating the diversion patterns into the LSTM model results in an improvement of 3.64% in Accuracy, 7.15% in Precision, and 1.34% in Recall. Results from the four mixed logit models indicate significant differences in factors contributing to traffic conflicts within each diversion pattern. For instance, increasing the speed difference between the target vehicle and the right preceding vehicle benefits traffic conflict during acceleration diversions but decreases the likelihood of traffic conflicts during deceleration diversions. These results can help traffic engineers propose individualized solutions to reduce unsafe diversion behavior.

Comparing fatal crash risk factors by age and crash type by using machine learning techniques

This study aims to use machine learning methods to examine the causative factors of significant crashes, focusing on accident type and driver's age. In this study, a wide-ranging data set from Jeddah city is employed to look into various factors, such as whether the driver was male or female, where the vehicle was situated, the prevailing weather conditions, and the efficiency of four machine learning algorithms, specifically XGBoost, Catboost, LightGBM and RandomForest. The results show that the XGBoost Model (accuracy of 95.4%), the CatBoost model (94% accuracy), and the LightGBM model (94.9% accuracy) were superior to the random forest model with 89.1% accuracy. It is worth noting that the LightGBM had the highest accuracy of all models. This shows various subtle changes in models, illustrating the need for more analyses while assessing vehicle accidents. Machine learning is also a transforming tool in traffic safety analysis while providing vital guidelines for developing accurate traffic safety regulations.

Enhancing autonomous vehicle hyperawareness in busy traffic environments: A machine learning approach

As autonomous vehicles (AVs) advance from theory into practice, their safety and operational impacts are being more closely studied. This study aims to contribute to the ever-evolving algorithms used by AVs during travel in busy urban districts, as well as explore the potential utilization of AV sensor data to identify safety hazards to surrounding road users in real time. Accordingly, the study incorporates AV data collected from multiple cities in the United States to detect and categorize traffic conflicts that involve the source AVs, as well as conflicts that involve other surrounding road users. Then, a machine learning conflict prediction model is trained with Isolation Forest - Convolutional Neural Network - Long Short-Term Memory (IF-CNN-LSTM) layers. The model receives data in real time in the form of road user trajectories and headings to make an informed prediction of the potential frequency and severity of conflicts three seconds into the future. In addition, the transferability of the trained model to new data and locations is explored to understand the potential compromise in accuracy compared to the effort and cost of retraining. The results show that the proposed model is capable of predicting the possibility of conflict occurrence and conflict severity with high accuracy (sensitivity = 83.5 % and fallout = 11 %). The reported sensitivity of AV conflict prediction ranged between 89 % and 95 %, depending on conflict type, which outperforms most of the existing conflict prediction models. The model is also capable of predicting hazardous conflicts of surrounding road users in real time, with sensitivity values ranging between 82 % and 87 %, affirming the promising capabilities of onboard vehicle sensors in undertaking real-time safety applications. The model also retains good performance when transferred to different data, with the potential to retain nearly 97 % of the source model's performance if sufficient tuning data exists.