Prediction of Duration of Traffic Incidents by Hybrid Deep Learning Based on Multi-Source Incomplete Data

Abstract

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 accident duration prediction using multi-mode data and ensemble deep learning

“…Therefore the BiGRU-CNN has the highest F1 value of 0.527. This finding is consistent with Shang et al [ 33 ] who have reported the importance of deep learning algorithms in the duration prediction of traffic accidents by text data. In addition, this study confirms that the newly established structured features extracted from text data by the one-hot model substantially enhance the prediction effects of deep learning algorithms.…”

“…Similarly, Ji et al predicted the duration of highway accidents based on the text data associated with 969 traffic accident records and social network information, and an average mean absolute percentage error (MAPE) of less than 22% was obtained for accident durations within the range of 0–180 min [ 32 ]. Shang et al established a mixed deep learning model combining latent Dirichlet allocation-BiLSTM (LDA-BiLSTM) networks to predict the duration of traffic accidents based on an analysis of the text data associated with 78,317 traffic incidents [ 33 ]. However, the traffic accident durations considered in this work were limited to within 90 min, and far too little attention was given toward capturing multi-mode data via the fusion of structured data and text data.…”

A Novel Explanatory Tabular Neural Network to Predicting Traffic Incident Duration Using Traffic Safety Big Data

“…Some models combine the strengths of statistical models and machine learning models. Hybrid models [3,27,30] can be more accurate than either statistical models or machine learning models alone. One example of a hybrid model is the Bayesian network [31].…”

“…The term "duration" in the context of a traffic incident refers to the overall time period starting from when the incident takes place until the arrival of traffic police at the scene and the completion of handling. Generally, the duration of a traffic incident can be divided into the following stages [3], as illustrated in Figure 1 below.…”