Considering Trip Generation and Route Selection in Regression-Based Prediction of Traffic Volumes

Noshin Saiyara Ahmad, Purdue University

Abstract

In today’s fast-paced data-driven world, accumulating and organizing streams of highresolution information plays a vital role in numerous decision and design tasks. The transportation sector is a prime example of this. Fine-scale information on traffic exposure at specific observation periods is critical to the successful analysis of road safety. Annual Average Daily Traffic (AADT) and hourly traffic volumes represent essential statistics to predict crash risk under time-dependent conditions, such as, weather and seasonal traffic variations. State highway agencies including the Indiana Department of Transportation (INDOT) collect traffic count data using multiple permanent and coverage count stations. However, approximately ten percent of the local-administered road segments in Indiana are included in their database. To impute the missing data, predictive models that can accurately forecast AADT and consequently, hourly traffic volumes, will be of great value. To address this problem, this thesis proposes a methodology to predict traffic volumes in different classes of urban road segments in Indiana. Two sets of regression models have been developed: (1) AADT Estimation Model, and (2) Hourly Traffic Volume Model. These models include effects of spatial and temporal variations, land use, roadway characteristics and, previously-overlooked in such models, road network connectivity and route selection. These, in turn, address two important research questions: (1) how trips are generated and (2) how people choose routes. The spatial and temporal effects that were considered in the analysis are travel propensity, travel time excess index, road class, hour of day, day of week and seasonal variations. While travel propensity captures particulars of network connectivity and land-use characteristics in traffic analysis zones (TAZ), the travel time excess index accounts for commuters’ route-choice. The estimation results indicate that all these variables are strongly correlated with traffic volumes on considered roadways. Reasonable estimations of hourly traffic volumes on a network scale can be achieved using the proposed model. In addition to aiding safety management at disaggregate level, hourly traffic predictions can help highway agencies in other system-wide analysis where such traffic information is needed.

Degree

M.Sc.

Advisors

Tarko, Purdue University.

Subject Area

Artificial intelligence|Civil engineering|Economics|Land Use Planning|Management|Medicine|Occupational safety|Transportation

COinS