Estimating historical hourly traffic volumes via machine learning and vehicle probe data: A Maryland case study

Sekuła, Przemysław; Marković, Nikola; Laan, Zachary Vander; Sadabadi, Kaveh Farokhi

doi:10.1016/j.trc.2018.10.012

Cited by 45 publications

(54 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Many of these studies evaluated probe performance using travel time reliability measures, such as the 90th or 95th percentile of travel time, standard deviation, percentage of variation, buffer time index (BTI), planning time index (PTI), travel time index (TTI), frequency of congestion, failure rate (with respect to average), on-time arrival, misery index, congestion detection latency, count of congestion, congestion duration, reliability curve, hourly traffic volume, congested hours, etc. (Aliari and Haghani 2012;Araghi et al 2015;Belzowski et al 2014;Cookson and Pishue 2016;Day et al 2015;FHWA 2017;Gong and Fan 2017;Lomax et al 2003a, b;Mcleod et al 2012;MoDOT 2017;Peniati 2004;Pu 2012;Remias et al 2013;Schrank et al 2012Schrank et al , 2015Sekuła et al 2017;Turner 2013;Venkatanarayana 2017;WSDOT, 2013WSDOT, , 2014Zheng et al 2018). An overview of these studies and the performance measures used to evaluate the reliability of probe data is provided in Table 1.…”

Section: Wide Area Probe Datamentioning

confidence: 99%

Improving Probe-Based Congestion Performance Metrics Accuracy by Using Change Point Detection

Ahsani

Sharma

Hegde

et al. 2020

J. Big Data Anal. Transp.

View full text Add to dashboard Cite

Probe-based speed data provide great value to agencies; especially, in areas which are not feasibly covered by traffic sensors. However, as with sensors, probe data are not without nuance and issues like latency prevent alignment between calculated metrics by data source. In recent years, there has been a strong impetus on using data-driven decision making. Data-driven insights have become critical for smart mobility. To support data-driven decision making, Federal Highway Administration has procured probe data feeds and provides free access to state and local agencies as National Performance Measures Research dataset (NPRMDS). In addition to the NPRMDS, several state agencies subscribe to a paid probe data provider for obtaining real-time streams of high-resolution probe data. These datasets are used to generate nationwide urban mobility reports as well as reports focusing on certain jurisdiction. These mobility reports are integrated in several Transportation System Management and Operations (TSMO) plans which are often used to drive several resource allocation projects. This paper examines accuracy of methodology used to derive two frequently used performance measures in the mobility reports; namely, number of congested hours and number of congestion events. An improve methodology is then proposed to find accurate estimates for number of congested hours and number of congested incidents.

show abstract

Section: Wide Area Probe Datamentioning

confidence: 99%

Improving Probe-Based Congestion Performance Metrics Accuracy by Using Change Point Detection

Ahsani

Sharma

Hegde

et al. 2020

J. Big Data Anal. Transp.

View full text Add to dashboard Cite

show abstract

“…Task 2-3 involved exploration of the base data set for freight applications. UMD completed this task, and the information gained was summarized in a published journal article (Sekula et al, 2018). NREL participated by reviewing UMD's work and attending a webinar on this topic.…”

Section: Task 2-3 Freight (Origin and Destination) Data And Practicesmentioning

confidence: 99%

Real-Time Volume and Turning Movements from Probe Data (CRADA CRD-16-614 Final Report)

Young

Garikapati

Hou

2020

View full text Add to dashboard Cite

show abstract

“…Machine learning has proven its ability to provide accurate estimates for different traffic characteristics [23,24,25,26,27,28]. Traffic speed and density have been estimated using an artificial neural network (ANN) model [23].…”

Section: Related Workmentioning

confidence: 99%

“…Results indicated higher accuracy from the support vector machine algorithm than the k-nearest neighbor classification algorithm. Estimating hourly traffic volumes between sensors was addressed using an NN model in the Maryland highway network [27], deploying both probe vehicles and automatic traffic recording station data to construct the NN model. A comparison was also made between linear regression, k-nearest neighbor, support vector machine with linear kernel, random forest, and NN models, concluding that the NN model performed the best.…”

Section: Related Workmentioning

confidence: 99%

Developing a Neural–Kalman Filtering Approach for Estimating Traffic Stream Density Using Probe Vehicle Data

Aljamal

Abdelghaffar

Rakha

2019

Sensors

View full text Add to dashboard Cite

This paper presents a novel model for estimating the number of vehicles along signalized approaches. The proposed estimation algorithm utilizes the adaptive Kalman filter (AKF) to produce reliable traffic vehicle count estimates, considering real-time estimates of the system noise characteristics. The AKF utilizes only real-time probe vehicle data. The AKF is demonstrated to outperform the traditional Kalman filter, reducing the prediction error by up to 29%. In addition, the paper introduces a novel approach that combines the AKF with a neural network (AKFNN) to enhance the vehicle count estimates, where the neural network is employed to estimate the probe vehicles’ market penetration rate. Results indicate that the accuracy of vehicle count estimates is significantly improved using the AKFNN approach (by up to 26%) over the AKF. Moreover, the paper investigates the sensitivity of the proposed AKF model to the initial conditions, such as the initial estimate of vehicle counts, initial mean estimate of the state system, and the initial covariance of the state estimate. The results demonstrate that the AKF is sensitive to the initial conditions. More accurate estimates could be achieved if the initial conditions are appropriately selected. In conclusion, the proposed AKF is more accurate than the traditional Kalman filter. Finally, the AKFNN approach is more accurate than the AKF and the traditional Kalman filter since the AKFNN uses more accurate values of the probe vehicle market penetration rate.

show abstract

Estimating historical hourly traffic volumes via machine learning and vehicle probe data: A Maryland case study

Cited by 45 publications

References 15 publications

Improving Probe-Based Congestion Performance Metrics Accuracy by Using Change Point Detection

Improving Probe-Based Congestion Performance Metrics Accuracy by Using Change Point Detection

Real-Time Volume and Turning Movements from Probe Data (CRADA CRD-16-614 Final Report)

Developing a Neural–Kalman Filtering Approach for Estimating Traffic Stream Density Using Probe Vehicle Data

Contact Info

Product

Resources

About