DxNAT — Deep neural networks for explaining non-recurring traffic congestion

Abstract

Non-recurring traffic congestion is caused by temporary disruptions, such as accidents, sports games, adverse weather, etc. We use data related to real-time traffic speed, jam factors (a traffic congestion indicator), and events collected over a year from Nashville, TN to train a multi-layered deep neural network. The traffic dataset contains over 900 million data records. The network is thereafter used to classify the realtime data and identify anomalous operations. Compared with traditional approaches of using statistical or machine learning techniques, our model reaches an accuracy of 98.73 percent when identifying traffic congestion caused by football games. Our approach first encodes the traffic across a region as a scaled image. After that the image data from different timestamps is fused with event-and time-related data. Then a crossover operator is used as a data augmentation method to generate training datasets with more balanced classes. Finally, we use the receiver operating characteristic (ROC) analysis to tune the sensitivity of the classifier. We present the analysis of the training time and the inference time separately.

Cooperative Evaluation of the Cause of Urban Traffic Congestion via Connected Vehicles

“…Unlike the approaches proposed in the literature, we deal with uncertainties via cooperation between vehicles on the road segment before data fusion can take place. Moreover, data mining techniques such as clustering, association, classification, have been applied in VANET to extract useful patterns and information [11]. Particularly, we explore collected data between CVs for extracting relationships via data mining techniques.…”

HUAD: Hierarchical Urban Anomaly Detection Based on Spatio-Temporal Data

“…By analyzing the spatial and temporal patterns of traffic accident frequency, Ren et al presented the spatio-temporal correlation of traffic accidents [15]. Sun et al propose a deep neural network model based on spatio-temporal data to identify non-recurring traffic congestion and explain its causes [16]. Wang et al develop a two-stage method to effectively detect traffic anomalies from GPS snippets, thus solving the problem of noise and sparsity of GPS fragments collected by vehicles [17].…”

Real-time deep reinforcement learning based vehicle navigation

“…NRC is defined as the congestion made by unexpected events, such as construction work, inclement weather, accidents, and special events Hall (1993). Unsurprisingly, NRC accounts for a larger proportion of traffic delays in urban areas comparing to the RC due to its unpredictable nature Sun et al (2017). There are three categories of methods proposed to tackle the NRC problem: (1) detecting and predicting traffic congestions by utilizing both the historical and real-time sensor data Zygouras et al (2015); Ghafouri et al (2017); (2) optimizing traffic signal control and management Wen (2008); Mousavi et al (2017a); and (3) vehicle routing and navigation optimization Ritzinger et al (2016); Jabbarpour et al (2018); Okulewicz and Mańdziuk (2017); Abdulkader et al (2015).…”

“…Karlaftis et al conducted an overview comparing statistical methods with neural networks in transportation-related research and it demonstrated that solutions based on deep reinforcement learning are very promising Karlaftis and Vlahogianni (2011). Most research projects in this area apply deep learning for traffic prediction Lv et al (2015); Polson and Sokolov (2016) or accident prediction Ren et al (2018); Sun et al (2017) to detect traffic congestions in advance. For traffic prediction, Lv et al proposed a deep learning-based traffic flow prediction method by using a stacked auto-encoder model to learn generic traffic flow features Lv et al (2015).…”

“…Ren et al Ren et al (2018) collected traffic accident data and analyzed the spatial and temporal patterns of the traffic accident frequency for the accident risk pre-dictor. Sun et al Sun et al (2017) proposed a deep neural network DxNAT to identify non-recurring traffic congestion by converting traffic data in Traffic Message Channel (TMC) format to an image, and use a convolutional neural network (CNN) to identify non-recurring traffic anomalies. Similarly, although the papers above can identify the non-recurring traffic congestion, a decision-making method that aims to help the user to avoid those traffic congestions is missing.…”