Predicting Queued Traffic Conditions in Work Zones Using Multisource Probe Data

Abstract: Detecting and predicting slow or stopped traffic conditions is essential for implementing policies to improve the level of service in work zones. This study investigates the potential of using various probe data to predict queueing conditions in work zones defined by speeds of 15 mph or less. The study employed segment speed data, speed from sensors, and connected vehicle speed data from five work zones using data in June 2022 for its analysis. We developed and compared the performance of several deep learning models, including Feedforward Neural Networks (FFNN), Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM) networks, and Convolutional Neural Networks (CNN) to predict queued conditions in work zones. The data was preprocessed to include time-based features (hour and day of the week ) and encoded categorical variables (work zone name and direction). The models exhibited varying degrees of accuracy across these zones, experiencing higher prediction accuracies (up to 95% F1-score). The findings reveal that integrating multiple data sources significantly enhances the predictive accuracy of queueing conditions compared to models relying on a single probe data source. Also, the results demonstrate the importance of utilizing various probe data sources and deep learning techniques to predict queueing and develop measures to improve traffic flow in work zones.