Deep representation of imbalanced spatio‐temporal traffic flow data for traffic accident detection

Abstract

Automatic detection of traffic accidents has a crucial effect on improving transportation, public safety, and path planning. Many lives can be saved by the consequent decrease in the time between when the accidents occur and when rescue teams are dispatched, and much travelling time can be saved by notifying drivers to select alternative routes. This problem is challenging mainly because of the rareness of accidents and spatial heterogeneity of the environment. This paper studies deep representation of loop detector data using long‐short term memory (LSTM) network for automatic detection of freeway accidents. The LSTM‐based framework increases class separability in the encoded feature space while reducing the dimension of data. The experiments on real accident and loop detector data collected from the Twin Cities Metro freeways of Minnesota demonstrate that deep representation of traffic flow data using LSTM network has the potential to detect freeway accidents in less than 18 min with a true positive rate of 0.71 and a false positive rate of 0.25 which outperforms other competing methods in the same arrangement.

Predicting Highway Risk Event with Trajectory Data: A Joint Approach of Traffic Flow and Vehicle Kinematics

“…The Synthetic Minority Oversampling Technique (SMOTE) is a traditional oversampling strategy that balances the dataset by generating synthetic samples from minority classes using the k-nearest neighbors algorithm and linear interpolation [60]. This technique has been extensively used in real-time crash prediction studies [61,62]. Conversely, the undersampling strategy balances the class proportions by removing samples from the majority class [17,42].…”

Comprehensive analysis of UK AADF traffic dataset set within four geographical regions of England

“…After that, they realized traffic flow detection by developing a real‐time vehicle tracking counter by combining these two algorithms. Mehrannia et al (2023) also employed a deep learning algorithm in their research on traffic accident detection. Based on the real traffic flow and accident data from the Twin Cities Metro freeways of Minnesota, they utilized long‐short term memory (LSTM) network to extract the features of traffic flow data and train their model to label the data samples as crash or non‐crash.…”

Prediction of the Road Accidents Severity Level: Case of Saint-Petersburg and Leningrad Oblast

“…In order to see how prediction accuracy changes depending on the usage of the SMOTE algorithm and sample, we estimate logit models using subsamples for Saint -Petersburg and Leningrad oblast and combined sample. In addition, for each case, we conduct modeling using both initial data and oversampled data (Mehrannia et al, 2023;Mostafa, Salem, and Habashyis, 2022;Shirwaikar et al, 2022;Sobhana et al, 2022).…”

“…To address this issue, they employ various methods before modeling, such as the SMOTE method, clustering analysis, and data undersampling techniques, among others. An example of using the SMOTE method can be the works (Mostafa, Salem, and Habashyis, 2022) and (Mehrannia et al, 2023), where using this method the sample was balanced, and further model construction was carried out. The method of synthetic oversampling of the minority was also used in the works (Shirwaikar et al, 2022) and (Sobhana et al, 2022) devoted to the analysis of the road accident severity levels.…”