Accident Dataset Research Articles

Can Historical Accident Data Improve Sustainable Urban Traffic Safety? A Predictive Modeling Study

Traffic safety is a critical factor for the sustainable development of urban transportation systems. This study investigates the impact of historical accident information on the prediction of future traffic accident risks, as well as the interaction between this information and other features, such as driver violations and vehicle attributes. Using a comprehensive dataset of traffic accidents involving passenger vehicles in a western Chinese city, we developed two predictive models: Model 1, which is based on vehicle information and driver violations, and Model 2, which integrates historical accident data. The results indicate that the inclusion of historical accident information significantly enhances the predictive performance of the model, particularly in terms of AUC (Area Under the Curve) and AP (Average Precision) values. Furthermore, through feature importance analysis and SHAP (SHapley Additive exPlanations) value evaluation, this study reveals the interaction effects between historical accident data and other features, and how these interactions influence model decisions. The findings suggest that historical accident data play a positive role in predicting future accident risk, with varying effects on risk mitigation. These insights provide a scientific basis for developing strategies to ensure the sustainable development of urban transportation systems.

Estimating casualties from urban fires: A focus on building and urban environment information

This study developed two prediction models for urban fire occurrence and related casualties via a fire accident dataset from Seoul, South Korea, from 2017 to 2021. Our models exhibit improved predictive performance by incorporating built environment features, such as building characteristics and the urban context, alongside weather and demographic data. This approach showed improved predictive performance suitable for public health implementation. Compared with the weather- and demographic-only models, our models had an 18.1 % greater fire occurrence prediction accuracy and a 10.4 % greater casualty prediction accuracy. Major variables affecting fire occurrence include building characteristics, e.g., the floor area ratio (FAR), building age, and commercial building number. Important features affecting casualty occurrence include demographic aspects, e.g., income level and weather, and network-based features, e.g., road connectivity and fire station proximity. These findings suggest that fire prevention strategies and fire casualty prevention strategies may need to differ. Furthermore, we identify high-risk zones by conducting spatial analysis and fire risk and casualty prediction on all buildings by applying our models to Seoul's Gangnam District. These contributions can promote safe and healthy urban environments by improving fire risk prediction accuracy and providing important insights into urban planning for appropriate urban fire accident response and prevention.

An accident severity prediction framework with consider features interaction

Accident severity prediction is increasingly important for preventing and reducing traffic accident losses. However, many studies always ignore the complex relationship between features during feature selection. To improve the prediction accuracy of the accident severity prediction model, this paper considers the interactions between multiple features. First, the feature selection algorithm of Recursive Feature Elimination with Cross-Validation (RFECV) is improved by using Shapley Additive explanations (SHAP) as the feature importance assessment metric. Then, to decrease the time expense of manually finding hyperparameters of the model, the Hunter Prey Optimization (HPO) algorithm is introduced and logistic mapping together with stochastic perturbation is added to it, which makes it easier to skip out of the partial optimum during the optimization search. Finally, the improved HPO algorithm is used to optimize the hyperparameters of the CatBoost model. The US traffic accident dataset is introduced for the validity of the proposed framework. Experimental results show that the proposed framework achieves a prediction accuracy of 96.63%, which is better than other state-of-the-art methods. The high accuracy of the prediction model can help decision-makers develop more rational transportation policies, and this study also proposes some traffic management measures based on the selected features.

Triplet-set feature proximity learning for video anomaly detection

The identification of anomalies in videos is a particularly complex visual challenge, given the wide variety of potential real-world events. To address this issue, our paper introduces a unique approach for detecting divergent behavior in surveillance videos, utilizing triplet-loss for video anomaly detection. Our method involves selecting a triplet set of video segments from normal (n) and abnormal (a) data points for deep feature learning. We begin by creating a database of triplet sets of two types: a-a-n and n-n-a. By computing a triplet loss, we model the proximity between n-n chunks and the distance between ‘a’ chunks from the n-n ones. Additionally, we train the deep network to model the closeness of a-a chunks and the divergent behavior of ‘n’ from the a-a chunks.The model acquired in the initial stage can be viewed as a prior, which is subsequently employed for modeling normality. As a result, our method can leverage the advantages of both straightforward classification and normality modeling-based techniques. We also present a data selection mechanism for the efficient generation of triplet sets. Furthermore, we introduce a novel video anomaly dataset, AnoVIL, designed for human-centric anomaly detection. Our proposed method is assessed using the UCF-Crime dataset encompassing all 13 categories, the IIT-H accident dataset, and AnoVIL. The experimental findings demonstrate that our method surpasses the current state-of-the-art approaches. We conduct further evaluations of the performance, considering various configurations such as cross-dataset evaluation, loss functions, siamese structure, and embedding size. Additionally, an ablation study is carried out across different settings to provide insights into our proposed method.

Towards efficient and accurate prediction of freeway accident severity using two-level fuzzy comprehensive evaluation

Accurately predicting freeway accident severity is crucial for accident prevention, road safety, and emergency rescue services in intelligent freeway systems. However, current research lacks the required precision, hindering the effective implementation of freeway rescue. In this paper, we efficiently address this challenge by categorizing influencing factors into two levels: human and non-human, further subdivided into 6 and 36 categories, respectively. Furthermore, based on the above factors, an efficient and accurate Freeway Accident Severity Prediction (FASP) method is developed by using the two-level fuzzy comprehensive evaluation. The factor and evaluation sets are determined by calculating the fuzzy evaluation matrix of a single factor. The weight matrix is calculated through the entropy method to compute the final evaluation matrix. Based on the maximum membership principle, the severity of the freeway accident is predicted. Finally, based on the experiments conducted with the traffic accident datasets in China and the US, it is shown that FASP is able to accurately predict the severity of freeway traffic accidents with thorough considerations and low computational cost. It is noted that FASP is the first attempt to achieve freeway accident severity prediction using the two-level fuzzy comprehensive evaluation method to the best of our knowledge.

A new class of zero-truncated counting models and its application

Count data is a type of data derived from the number of times an event occurs per unit of time, and zero-truncated count data refers to count data without zero, which often appears in various fields. In this paper, a new zero-truncated Bell (ZTBell) distribution is proposed on the basis of Bell distribution. We studied its statistical properties, exploring methods such as maximum likelihood estimation (MLE), expectation–maximization (EM) algorithm, and minimization–maximization (MM) algorithm for parameter estimation, as well as conducting likelihood ratio tests. In addition, we used the Bootstrap method to calculate the standard errors and confidence intervals of the parameters. The simulation results found that all of the MLE, MM algorithm and EM algorithm are effective. And, as the sample size increases, the estimates of the parameters are closer to the true values and the root mean square error is smaller. Finally, applying the model to a set of factory accident data, we found that the ZTBell distribution fits better than the other models and is close to the fitting results of the zero-truncated generalized Poisson distribution. But ZTBell distribution has only one parameter, so it’s even simpler compared to the latter. Therefore, the ZTBell distribution can be a good alternative to other zero-truncated distributions, which provides more options available for statistical analysis in this domain.

Accident Detection and Nearby Patrol Alert System Using AI &ML

Abstract: Intelligent communities are utilizing different creative ideas to improve the quality of human life. Due to fast growing sizes of our cities, need of travelling is constantly increasing, which in turn has increased count of vehicles on the roads. Increasing number of vehicles on the roads has brought about numerous difficulties for Street Traffic Management Authorities. Amongst different traffic related issues, road accidents are something worth giving attention to and have to be on the priority list. This project describes various automatic road accident detection techniques, which automatically detect accidents using surveillance videos in real-time and intimates the event scenario to control room and the main uniqueness of the system is it automatically estimates the crash value of the incident and intimates to the user. The proposed method assumes that traffic accident events are described by visual features occurring through a temporal way. Therefore, a visual features extraction phase, followed by a temporary pattern identification, compose the model architecture. The visual and temporal features are learned in the training phase through convolution and recurrent layers using built-from-scratch and public datasets. An accuracy of 98% is achieved in the detection of accidents in public traffic accident datasets, showing a high capacity in detection independent of the road structure.

Incorporation of a global perspective into data-driven analysis of maritime collision accident risk

Ship collision accidents are one of the most frequent accident types in global maritime transportation. Nevertheless, conducting an in-depth analysis of collision prevention poses a formidable challenge due to the constraints of limited Risk Influential Factors (RIFs) and available datasets. This paper aims to incorporate a global perspective into a new data-driven risk model, scrutinize the root causes of collision accidents, and advance measures for their mitigation. Additionally, it seeks to analyze the spatial distribution and conduct a comprehensive comparative study on collision characteristics for both pre- and post-COVID-19, utilizing the real accident dataset collected from two reputable organizations: Global Integrated Shipping Information System (GISIS) and Lloyd's Register Fairplay (LRF). The research findings and implications encompass several crucial aspects: 1) the constructed model demonstrates its reliability and accuracy in predicting collision accidents, as evident from its prediction performance and various scenario analysis; 2) the most hazardous voyage segment for collision accidents is identified to provide valuable guidance to different stakeholders; and 3) the hierarchical significance of various ship types in the context of collision accident is highlighted regarding the most probable scenario for collision occurrences; 4) During the pandemic, the rise in collision probabilities, particularly involving older vessels and bulk carriers, implies heightened operational challenges or maintenance issues for these ship types; (5) The prominence of favorable and adverse sea conditions in collision reports underscores the significant influence of weather on accidents during the pandemic. These findings and implications help enhance safety protocols, ultimately reducing the frequency of collision accidents in the global maritime domain.

Applying a two-step cluster algorithm in traffic accident data analysis

Cluster analysis is often employed as the initial stage in organizing heterogeneous data into homogeneous groups. Choosing an effective clustering approach and an ideal number of clusters in a traffic accident dataset might be complex and challenging. This study aims to evaluate the effectiveness of k-means and two-step cluster methods. Subsequently, the two-step cluster method and GIS are applied to analyze the traffic accident datasets from 2015 to 2017 in Hanoi, Vietnam. First, according to the Silhouette score, the two-step cluster method achieved a higher score of 0.563, while the k-means method scored 0.341. A higher Silhouette score indicates more well-defined clusters. Second, the research suggests combining the two-step cluster method with GIS for analyzing traffic accident datasets. The outcome identifies five typical types of accidents in Hanoi. In addition, the locations of various accident types were visually illustrated on a map, enabling traffic officials to recommend precise and urgent countermeasures. Importantly, the clustering results reveal that the two-step cluster method exhibits a significantly higher rate of homogeneous data in the clusters compared to the k-means method. This study demonstrates that the two-step cluster method is not only more effective than the k-means method in terms of clustering ability but also in data pre-processing. The study's results enable authorities to gain a more detailed understanding of typical traffic accident patterns in Hanoi. Besides, the employed methods could potentially be applied to other regions, providing an additional avenue for analysis

Impact of COVID-19 on road crashes in Thailand

The main goal of this study is to investigate the impact of COVID-19 on road crashes in Thailand using time series and interrupted time series analysis. To achieve the goal, road crash data from the Department of Highway (DOH), which includes total crashes, single vehicle crashes, fatalities, fatal crashes, speeding crashes, and drunk driving crashes, was obtained to conduct Seasonal Autoregressive Integrated Moving Average (SARIMA) time series models and Interrupted Time Series (ITS) models. SARIMA models were applied to forecast the number of crashes in the absence of COVID-19 then compare them to the observed values to identify the difference. The impact of a policy change aimed at addressing the spread of COVID-19 was assessed using ITS models on a time series accident dataset. The goal was to ascertain if the intervention had a meaningful and causative impact on the outcome. The result showed that the first wave of COVID-19 caused a significant reduction in all road crash indicators instead of skyrocketing to a peak. After releasing the lockdown measures from the first wave of spreading, an increase was found in all of the crash indicators as well. However, the third wave of COVID-19, which lasted longest for nearly 7 months, also caused a decrease in the number of crashes, but not as much as the first wave of the outbreak. Moreover, the result from the interrupted time series also revealed that curfews and the closure of entertainment places are associated with a significant decrease in the number of speeding crashes and drunk driving crashes from 10 p.m. to 4 a.m., respectively. It can be observed that the COVID-19 countermeasures, such as curfews and bans on the sales of alcoholic beverages, led to a drop in the number of speeding and drunk driving crashes.

Deciphering spatial heterogeneity of maritime accidents considering impact scale variations

ABSTRACT Ensuring maritime safety has ascended as a preeminent concern within the global maritime sector. Understanding how factors affect maritime accidents’ consequences in different water areas would be of great benefit to preventing the occurrence or reducing the consequences. This study thus employed a multi-scale geographically weighted regression (MGWR) model on the accident dataset from Fujian waters in the East China Sea, to quantify the influences of different factors as well as the spatial heterogeneity in the effects of key factors on maritime accident consequence. The performances of MGWR are compared with multiple linear regression (MLR) and GWR. As expected, MGWR outperforms the other two models in terms of its ability to clearly capture the unobserved spatial heterogeneity in the effects of factors. Results reveal notably distinct influences of some factors on maritime accident consequences in different locations. An intuitive indication by MGWR is that approximately 50% of the accidents present positive coefficients of good visibility while other locations are negative, which are failed to recognize by MLR. The outcomes provide insights for making appropriate safety countermeasures and policies customized for different water areas.

Hazardous traffic scenarios for motorcyclists in Indonesia: a comprehensive insight from police accident data and self-reports

Motorcycle safety remains a concern in low- and middle-income countries. This study addresses this issue by identifying hazardous scenarios for motorcyclists in Indonesia. We conducted a two-step cluster analysis and injury analysis to examine motorcycle accidents based on the police accident dataset (2020–2021) of Brebes Regency, Indonesia. We integrated the findings with accident self-reports from 104 young motorcyclists using a joint display to obtain a more comprehensive insight. As a result, we identified four hazardous traffic scenarios: motorcycle-to-vehicle collisions on median roads, motorcycle-to-vehicle collisions on non-median roads, motorcycle-to-pedestrian collisions, and single-motorcycle collisions. We suggest countermeasures for each scenario and propose a public transport policy as a safer mobility solution. Applying a two-step cluster analysis on accident data and integrating the findings of accident data and self-report analysis proved beneficial in this study. Therefore, we encourage the use of this novel approach in future studies.

BConvLSTM: a deep learning-based technique for severity prediction of a traffic crash

Predicting the severity of crashes has become a significant issue in research on road accidents. Traffic accident severity prediction is essential for protecting vulnerable road users and preventing traffic accidents. For practitioners to identify significant risk variables and set appropriate countermeasures in place, explainability of the forecast is also essential. Most previous research ignores the severity of property loss caused by traffic accidents and cannot differentiate between different levels of fatalities and property loss severity. Additionally, while an understandable structure of deep neural networks (DNN) is significantly lacking in existing works, understanding traditional systems is quite simple. An inability to use structural data when describing forecasting and the many attempts to incorporate neural networks afflict the absence of hidden layers. We propose a Deep Learning (DL) framework for forecasting traffic crash severity to overcome the accident severity prediction. It has three steps to process. Initially, collected input data are cleaned. Data cleaning is performed in a preprocessing step. We conduct experiments on two datasets, A Countrywide (US) Traffic Accident Dataset and UK Road Accident Dataset. The outcomes of the experiments demonstrate that the proposed technique outperformed other approaches and produced the best accuracy.

Advancing urban traffic accident forecasting through sparse spatio-temporal dynamic learning

Traffic accidents have emerged as one of the most public health safety matters, raising concerns from both the public and urban administrators. The ability to accurately predict traffic accident not only supports the governmental decision-making in advance but also enhances public confidence in safety measures. However, the efficacy of traditional spatio-temporal prediction models are compromised by the skewed distributions and sparse labeling of accident data. To this end, we propose a Sparse Spatio-Temporal Dynamic Hypergraph Learning (SST-DHL) framework that captures higher-order dependencies in sparse traffic accidents by combining hypergraph learning and self-supervised learning. The SST-DHL model incorporates a multi-view spatiotemporal convolution block to capture local correlations and semantics of traffic accidents, a cross-regional dynamic hypergraph learning model to identify global spatiotemporal dependencies, and a two-supervised self-learning paradigm to capture both local and global spatiotemporal patterns. Through experimentation on New York City and London accident datasets, we demonstrate that our proposed SST-DHL exhibits significant improvements compared to optimal baseline models at different sparsity levels. Additionally, it offers enhanced interpretability of results by elucidating complex spatio-temporal dependencies among various traffic accident instances. Our study demonstrates the effectiveness of the SST-DHL framework in accurately predicting traffic accidents, thereby enhancing public safety and trust.

Severitys prediction of car accidents in PA and model comparison

This study aims to provide an analytical and predictive framework for understanding and forecasting the probabilities of varying severities of traffic accidents in Pennsylvania, based on the US Accident dataset dated up to March 23. Initially, the dataset and variable descriptions are provided for comprehensive understanding. During the data preprocessing phase, variables undergo a thorough examination for missing values, and suitable imputation methods are selected for data completeness. Further, feature selection and data cleansing are executed to prepare the dataset for model training. The study predominantly utilizes three machine learning algorithms, Logistic Regression, Random Forest, and Support Vector Machine, to construct predictive models. The performance of these models is meticulously evaluated for accuracy and compared through specific data sampling techniques. Overfitting checks, feature importance elucidation, and in-depth discussions on model performance variations are also included. By navigating through this multifaceted analysis, the study aims to shed light on the strengths and weaknesses of different modeling approaches for this particular problem context, thereby providing valuable insights for future research endeavors.

Traffic Accident Severity Prediction Based on Data Cleaning and Machine Learning (Random Forest / Xgboost)

Traffic accidents have increasingly become a global concern, significantly affecting lives and economic sustainability. The US Accidents dataset from 2016 to 2023 provides an extensive record of accidents across the United States, containing detailed data and environmental data of these accident. This study aims to harness the potential of this rich database to predict severity level of accidents. Our research predominantly revolved around meticulous data cleaning, ensuring that the dataset's integrity was uncompromised. After preprocessing, the cleaned data was subjected to sophisticated Machine Learning techniques, primarily focusing on the Random Forest and XGBoost algorithms. These models were chosen due to their renowned capability in handling complex datasets and rendering accurate predictions, especially in scenarios laden with multiple variables. Upon application, the models demonstrated impressive efficacy. To validate the reliability and performance of our models, we employed the confusion matrix. This tool provided a clear visualization of the models' accuracy, revealing true positives, false negatives, and other crucial metrics. Furthermore, to enhance prediction outcomes, the Voting Classifier was implemented, combining the strengths of our primary models and consequently elevating the overall accuracy. The Random Forest algorithm exhibited substantial precision, while XGBoost further enhanced prediction accuracy. These findings underline the significant role of advanced data analytics and Machine Learning in comprehending traffic accident dynamics. In conclusion, our study emphasizes that leveraging state-of-the-art Machine Learning techniques on well-curated datasets can substantially improve our understanding and prediction of traffic accident severity. Such insights pave the way for the development of more effective preventive measures and safety protocols, aiming for a safer traffic environment in the future.

Using Multilayer Perceptron Neural Network to Assess the Critical Factors of Traffic Accidents

This study is based on the traffic accident data of Taoyuan City from the government's open data. The study compiled the data set of traffic accidents in Taiwan from 2012 to 2017, and six classifiers were applied to evaluate the effectiveness of traffic accident prediction with the number of injuries as the prediction target. In order to verify the classifier's stability, cross-validation was used to evaluate the model during the training process, and the multilayer perceptron neural network (MLPNN) classifier performed best in testing the dataset's accuracy and evaluating the model's best performance. Then, a boosting ensemble learning approach and a combination of traffic accident factors improve the experiment's performance. According to this experiment, the results show that this study uses the Pearson Chi-square feature selection method to select important traffic factor combinations, and the boosting method indeed helps improve the effectiveness of the construction of the traffic accident model. Finally, the experimental results of the NN-MLP model have a correct rate of 77% and AUC is 78.7%. In constructing the model, it was found that the degree of injury, the part of the vehicle hit, the type of accident, the leading cause, the type of vehicle, and the period of the accident were the main factors causing dangerous traffic accidents. Doi: 10.28991/HIJ-2024-05-01-012 Full Text: PDF

Learning spatial patterns and temporal dependencies for traffic accident severity prediction: A deep learning approach

Traffic accidents have a substantial impact on human life and property, resulting in millions of injuries every year. To ensure road safety and enhance the research in this direction, it is necessary to develop methods that can efficiently predict and classify the accident severity. However, traffic accident datasets may contain a large number of features, making it challenging to extract relevant information and patterns from high-dimensional data. Moreover, traffic accidents may be influenced by multiple factors and temporal dependencies, leading to a dynamic impact of each factor on accident severity over time. To address these challenges, we propose a novel deep-learning approach for predicting traffic accident severity. Specifically, we first conduct a thorough data preprocessing step to clean the data and ensure its quality. Then, a Convolutional Neural Network (CNN) is introduced to extract spatial features and patterns from the high-dimensional data, followed by a Bidirectional Long Short-Term Memory network (BiLSTM) to capture the temporal dependencies between various factors that affect traffic accidents. We also implement attention mechanisms to weigh the importance of each feature in the prediction, thereby reducing the impact of noisy or irrelevant data. To evaluate the effectiveness of our approach, we conduct experiments on a real-world traffic accident dataset from two cities. The results demonstrate the practicality and effectiveness of our framework for traffic accident severity prediction, with potential to enhance road safety.

Improving Traffic Accident Severity Prediction Using Convoluted Features and Decision-Level Fusion of Models

Although there have been improvements in traffic safety measures, the frequency of traffic accidents continues to persist. Developing countries experience a significant impact from traffic accidents with respect to fatalities and property damage. Traffic accidents happen for multiple reasons, involving traffic conditions, driving violations, driver misjudgments, and so forth. Severe casualties may lead to fatalities; therefore, accident severity prediction might help reduce the chances of fatalities. This research makes use of a U.S. road accident dataset that contains the most relevant 32 factors related to accidents. For obtaining accurate prediction of traffic accident severity, this research proposes a solution based on an ensemble of random forest and support vector classifiers that is trained using deep convoluted features. Features are extracted from the road accident dataset using a convolutional neural network (CNN). The performance of models using original features and CNN features is analyzed that shows the superiority of convoluted features. Experimental results involving the use of several well-known machine learning models indicate that the proposed model can obtain an accuracy of 99.99% for traffic accident severity prediction. The efficacy of the proposed model is validated against existing state-of-the-art approaches.

Assessment of aviation accident datasets in severity prediction through machine learning

The importance and seriousness of civil aviation safety have become more noticeable due to the intricate advancement of air transportation. It is crucial to apply diverse datasets to evaluate and anticipate the degree of aviation safety. Data on the different causes of accidents can be analyzed and applied to predict and prevent potential accidents. The main goal of this research is to explore how Machine Learning (ML) techniques can be used to identify correlations and connections between various contributing factors in aviation accidents. It applies ML algorithms on datasets extracted from original data sourced from the National Transportation Safety Board (NTSB) of the United States, Transportation Safety Board (TSB) of Canada, and Australian Transport Safety Bureau (ATSB) aviation accident records spanning from January 2008 to January 2023. Initially, this study uses Artificial Neural Network (ANN), Decision Tree (DT), and Support Vector Machine (SVM) models to predict the civil aviation accident rates for the United States, Canada, and Australia data separately. Sequentially, by extracting Common Features (CF) and reclassifying the Common Sub-Features (CSF) of these three countries, a comprehensive model is created. Finally, it evaluates the differences and limitations of these countries' datasets. The results show that the separate models exhibit a higher level of prediction accuracy and lower errors compared to the common models and the best results (81% accuracy) were achieved through SVM.