Researchers in Texas Assessed ML Techniques for Freeway Management Applications.
Texas, United States
Leveraging Artificial Intelligence (AI) Techniques to Detect, Forecast, and Manage Freeway Congestion
Summary Information
Improving the quality and efficiency of the surface transportation system requires the capability to predict the onset and propagation of prolonged congestion, along with monitoring atypical events and estimating their evolution. Machine Learning (ML) methods present a unique opportunity to estimate congestion measures by utilizing data from agency-owned sensors, third-party providers, and big enterprise data. The objective of this study was to validate the reliability of commercial data sources for planning and operations and understand which ML models or algorithms were most suitable for addressing Texas Department of Transportation’s (TxDOT) needs based on desirable use cases and data availability.
- Adopt different ML models to predict congestion measures for rural and urban areas based on the performance differences. This study found better performance of rural freeway models as compared to urban ones and recognized the influence of data characteristics inherent to different land use type. Identifying and understanding these characteristics can enhance model accuracy and effectiveness.
- Integrate explainable ML methodologies to improve the interpretability of the results. Using explainable ML methodologies can help researchers understand the influence of various factors on freeway congestion. For example, techniques like Shapley Additive Explanations (SHAP) can highlight the role of key factors such as traffic volume, job density, k-factor, and road geometry.
- Import ML congestion management techniques from states which are already successful. California, Florida, and Minnesota already implement ML strategies such as K-means cluster analysis, regression analysis, and Random Forest.
- Feed ML models with continuous big data streams. Having real-time traffic speed and volume data combined with incident information constantly input into ML models can help create actionable predictions about near-term roadway performance.
- Make use of third-party services when appropriate. Large primary data vendors can pipeline their output to cloud platforms for data cleaning, management, and analysis.
