Statistical analysis to measure estimate accuracy

Public Roads, March-April, 2004 by Keri A. Funderburg

For UCF's report, The Impact of Real-Time and Predictive Traffic Information on Travelers' Behavior in the 1-4 Corridor, researchers performed statistical analyses to evaluate the short-term prediction system used to estimate the travel times posted at www.trafficinfo.org. The travel times were exclusively for trips along a 64-kilometer (40-mile) corridor of Interstate 4 in Orlando, FL.

Travel time estimates for the Florida Web site are based on data from 70 loop detector stations and closed-circuit television cameras. The estimates are calculated using two prediction models. The short-term model, which UCF analyzed in this report, uses real-time estimates to make predictions. The long-term model also provides estimates but uses travel time information based on historical data.

Analysis of the short-term model showed that the travel times posted on the Web were less accurate during heavy congestion and more accurate during lighter congestion--most likely due to random fluctuations in travel speed during heavy congestion. The analysis also showed that when predicting travel times in the near future (within 15 minutes), the system became increasingly inaccurate as the estimates moved further into the future. In addition, the analysis revealed that the prediction system had a slight tendency to underestimate travel times.

[ILLUSTRATION OMITTED]

Rigorous analysis and real-world data are the keys to evaluating travel time estimates and ensuring the accuracy of traveler information. For more information about FHWA's study, visit www.itsdocs.fhwa.dot.gov/JPODOCS/REPTS_TE/13867.html. To download UCF's report, go to www.dot.state.fl.us/research-center/Completed_Proj/Summary_TE/FDOT_BC355_03.pdf.