Sparse Estimation of Travel Time Distributions Using Gamma Kernels
The authors address two shortcomings in present travel time distribution estimation methods, which specifically apply to the case of congested stop-and-go traffic for which the probability densities tend to be multi-modal. The first shortcoming is related to the determination of the number of modes, which can change from one location in the network to another, as well as by time of day. The second one is the wide-spread use of mixtures of Gaussian probability densities, which can assign positive probabilities to negative travel times and offer too little flexibility because of their symmetric shape. These drawbacks of the existing approaches have been tackled in this paper through the use of a sparse kernel density estimation (KDE) technique using asymmetric Gamma kernels. The sparse modeling techniques have the additional capability to automatically infer the minimum number of modes from the data, thereby avoiding the need to predefine the number of mixture components. The use of asymmetric gamma kernels ensures nonnegative supports while also providing increased flexibility in the shapes of the distributions. Experimental results using high-dimensional simulated and real-world travel time data illustrate the efficacy of the proposed method, and further illustrate that Gamma kernels indeed outperform the classical Gaussian kernels in terms of estimation accuracy with as few elements as possible.
-
Supplemental Notes:
- This paper was sponsored by TRB committee AHB45 Standing Committee on Traffic Flow Theory and Characteristics.
-
Corporate Authors:
500 Fifth Street, NW
Washington, DC United States 20001 -
Authors:
- Dilip, Deepthi Mary
- Freris, Nikolaos M
- Jabari, Saif Eddin
-
Conference:
- Transportation Research Board 96th Annual Meeting
- Location: Washington DC, United States
- Date: 2017-1-8 to 2017-1-12
- Date: 2017
Language
- English
Media Info
- Media Type: Digital/other
- Features: Figures; References; Tables;
- Pagination: 13p
- Monograph Title: TRB 96th Annual Meeting Compendium of Papers
Subject/Index Terms
- TRT Terms: Distributions (Statistics); Estimation theory; Nonparametric analysis; Traffic congestion; Travel time
- Uncontrolled Terms: Kernel density estimators
- Subject Areas: Highways; Planning and Forecasting;
Filing Info
- Accession Number: 01628670
- Record Type: Publication
- Report/Paper Numbers: 17-02971
- Files: TRIS, TRB, ATRI
- Created Date: Mar 8 2017 9:07AM