Providing accurate travel time estimation is essential for trip planning and traffic operations, as traffic and congestion cost Americans significant time and money each year, with the worst costs and delays occurring in and around major cities. This study explored how to more accurately predict short-term travel time using machine learning techniques that account for various data sources, including loop detectors, probe vehicles, weather condition, geometry, roadway incidents, roadwork, special events, and sun glare. The proposed dynamic approach was tested and evaluated using "experienced travel times" on 10 highway segments in the Northeast Illinois highway network (i.e., adding up the travel times of the 10 consecutive links at the current time of each link starting from the origin).
- Focus on occupancy information from loop detectors to provide more accurate travel time prediction, with caution to malfunction of loop detectors. This study found that in comparison with other traffic variables collected by loop detectors, occupancy measured by loop detectors best predicted traffic conditions and travel times. However, the analysis also showed that low prediction accuracy of some highways was due to malfunctioning loop detectors and the uncertainty regarding the location of loop detectors. Raw data points from loop detectors were found to contain missing and erroneous records, caused by malfunctioning detectors, deteriorating pavement, or other reasons.
- Fuse traffic data from multiple sources to improve the accuracy of traffic prediction models. Raw loop detector data that is susceptible to missing and erroneous records should be supplemented with data from multiple sources. Data from multiple data sources, including occupancy, weather information, roadway incidents, special events, and roadwork could have a considerable impact on travel time prediction and was found to increase prediction accuracy of the models in this study.
- Minimize the weight assigned to sun glare data when building travel time prediction models. This study revealed that including sun glare as a variable did not improve the accuracy of the traffic predictions. Other weather variables accounted for slow downs associated with sun glare. Additionally, since sun glare typically occurs during rush hour where traffic speed is already reduced and therefore will not be significantly affected by sun glare.