Traffic accidents causing nonrecurrent congestion and road traffic injuries seriously affect public safety. It is helpful for traffic operation and management to predict the duration of traffic incidents. Most of the previous studies have been in a certain area with a single data source. This paper proposes a hybrid deep learning model based on multi-source incomplete data to predict the duration of countrywide traffic incidents in the U.S. The text data from the natural language description in the model were parsed by the latent Dirichlet allocation (LDA) topic model and input into the bidirectional long short-term memory (Bi-LSTM) and long short-term memory (LSTM) hybrid network together with sensor data for training. Compared with the four benchmark models and three state-of-the-art algorithms, the RMSE and MAE of the proposed method were the lowest. At the same time, the proposed model performed best for durations between 20 and 70 min. Finally, the data acquisition was defined as three phases, and a phased sequential prediction model was proposed under the condition of incomplete data. The results show that the model performance was better with the update of variables.

Traffic flow prediction plays an important role in intelligent transportation system (ITS). However, due to the randomness and complex periodicity of traffic flow data, traditional prediction models often fail to achieve good results. On the other hand, external disturbances or abnormal detectors will cause the collected traffic flow data to contain noise components, resulting in a decrease in prediction accuracy. In order to improve the accuracy of traffic flow prediction, this study proposes a mixed traffic flow prediction model VMD‐WD‐LSTM using variational mode decomposition (VMD), wavelet threshold denoising (WD), and long short‐term memory (LSTM) network. Firstly, we decompose the original traffic flow sequence into K components through VMD and determine the number of components K according to the sample entropy of different K values. Then, each component is denoised by wavelet threshold to obtain the denoised subsequence. Finally, LSTM is used to predict each subsequence, and the predicted values of each subsequence are combined into the final prediction results. In addition, the performance of the proposed model and the latest traffic flow prediction model is compared on the several well‐known public datasets. The empirical analysis shows that the proposed model not only has good prediction accuracy but also has superior robustness.

Automatic incident detection (AID) has always been one of the focus issues in the field of transportation. However, due to the contingency and randomness of traffic incident, traffic incident samples are scarce and far less than non-incident samples. Therefore, unlike other scenarios using large-scale deep networks, traffic incident detection tackle at small and imbalanced sample size. Imbalanced, small sample data sets, inappropriate and incomplete initial variable sets make the AID model insensitive to incident samples, resulting in unsatisfactory model performance (low detection rate or high false alarm rate). Therefore, a hybrid AID method (SASYNO-RF-RSKNN) is proposed using self-adaptive synthetic over-sampling, random forest and random subspace k nearest neighbor. First, the spatial-temporal and real-time characteristics of traffic stream are used for the selection of appropriate initial variables to construct a relatively complete set of initial variables. Second, the SASYNO oversampling method is used to expand the original imbalanced sample database, so that the number of minority class samples is consistent with the number of most class samples. Then, feature variables are selected from the initial variables using the RF algorithm. Finally, the RSKNN ensemble algorithm with feature variables as input is employed to detect traffic incident. In addition, six indexes are used to evaluate model performance, including accuracy (ACC), false alarm rate (FAR), detection rate (DR), precision, Matthews correlation coefficient (MCC) and F1-score. Simultaneously, we also designed horizontal and vertical contrast experiments, and the experimental results show that SASYNO-RF-RSKNN model has superior performance. It is worth mentioning that experiments are implemented on two real-world datasets. Most indexes of the proposed model are the best compared with other five excellent machine learning algorithms. On the whole, the proposed model has a dependable and high-performance for traffic incident detection.

As an important part of intelligent transportation systems, traffic state classification plays a vital role for traffic managers when formulating measures to alleviate traffic congestion. The proliferation of traffic data brings new opportunities for traffic state classification. In this paper, we propose a hybrid new traffic state classification method based on unsupervised clustering. Firstly, the k-medoids clustering algorithm is used to cluster the daily traffic speed data from multiple detection points in the selected area, and then the cluster-center detection points of the cluster with congestion are selected for further analysis. Then, the self-tuning spectral clustering algorithm is used to cluster the speed, flow, and occupancy data of the target detection point to obtain the traffic state classification results. Finally, several state-of-the-art methods are introduced for comparison, and the results show that performance of the proposed method are superior to comparable methods.