Incorporating Weather Information into Real-Time Speed Estimates: Comparison of Alternative Models

Abstract

Weather information is frequently requested by travelers. Prior literature indicates that inclement weather is one of the most important factors contributing to traffic congestion and crashes. This paper proposes a methodology to use real-time weather information to predict future speeds. The reason for doing so is to ultimately have the capability to disseminate weather-responsive travel time estimates to those requesting information. Using a stratified sampling technique, cases with different weather conditions (precipitation levels) were selected and a linear regression model (called the base model) and a statistical learning model [using support vector machines for regression (SVR)] were used to predict 30-min-ahead speeds. One of the major inputs into a weather-responsive short-term speed prediction method is weather forecasts; however, weather forecasts may themselves be inaccurate. The effects of such inaccuracies are assessed by means of simulations. The predictive accuracy of the SVR models show that statistical learning methods may be useful in bringing together streaming forecasted weather data and real-time information on downstream traffic conditions to enable travelers to make informed choices.