Project. Traffic forecasting: non-parametric regressions Completed

Principal investigator. Dr. William T. Scherer, University of Virginia, 804-982-2069, wts@virginia.edu

External contact. James Robinson, VDOT, 1401 E. Broad St., Richmond, VA 23219, 804-786-6677

Project objective. To improve short-term traffic forecasting by non-parametric regressions.

Project abstract. Nonparametric regression is a widely applicable case-based reasoning technique that has found recent application in forecasting short-term traffic flows. Often, Intelligent Transportation Systems (ITS) are merely reactive without the ability to predict future traffic conditions. We are investigating using k Nearest Neighbor (k-NN) approach to produce more accurate forecasting models. However, nonparametric regression techniques like k-NN are extremely resource-intensive, with a bulk of the processing time being spent to find similar past cases. To be applied in real-time ITS applications, the k-NN algorithm must be modified to reduce the overall processing time. We are investigating three areas for improvement: data management, data structures, and hardware issues. This research specifically focuses on using k-d trees within the k-NN algorithm to reduce its processing time. We will investigate the use of nonparametric regression and k-d trees with multiple, large data sets.

Tasks. 1) Description of the proposed forecasting approach. 2) Evaluation of the proposed technology. 3) Analysis of issues regarding the implementation of the proposed approach.

Milestones. This project will be completed by June 30, 2000.

Student involvement. One full-effort (half-time) GRA for the duration of the project.

Budget.   $70,000

Relation to other research. No immediate relationship.

Technology transfer. Papers in professional journals.

Potential benefits. Inaccurate traffic forecasting is a major impediment to useful traffic information systems. An effective new kind of modeling could have widespread impact on ATIS.

TRB keywords.  ITS, forecasting