Traffic forecast model predicts turn counts without impossible values
LTV-CTDNet: Compositional Turning Decomposition for Short-Term Turning-Movement Forecasting
Machine Learning
Summary
Predicting how cars turn at intersections helps manage traffic signals better. The authors developed a model called LTV-CTDNet that makes short-term forecasts of turning movements using recent traffic and time information. Their model always predicts realistic, nonnegative counts and keeps totals consistent, unlike some older models that sometimes gave impossible negative numbers. They tested it with data from multiple locations and found it to be accurate and reliable.
What this means in practice
- •For traffic management teams: Improve signal timing and corridor operations by integrating realistic short-term turning movement forecasts without negative or inconsistent values.
- •For urban planning teams: Estimate vehicle turning volumes accurately at intersections to better design traffic flow strategies.
Authors
Md Atiqur Rahman Mallick, Kamrul Hasan, Robert T. White
Abstract
Short-term turning-movement forecasts can support signal control and corridor operations, but unconstrained neural networks may produce physically impossible negative counts or outputs that are not explicitly tied to an approach-demand total. This study introduces the Linear Temporal-Variable Compositional Turning Decomposition Network (LTV-CTDNet), a forecasting framework designed to combine competitive accuracy with structurally admissible outputs. LTV-CTDNet was evaluated using seven months of 15-minute LiDAR observations from eight monitored corridor locations in Nashville, Tennessee. Its lightweight encoder combines recent turning-movement history, weekly time-slot embeddings, and location embeddings. The Compositional Turning Decomposition framework separately predicts nonnegative approach totals and within-approach turning proportions, then reconstructs movement forecasts from these components. Among the evaluated predefined configurations, LTV-CTDNet achieved a movement-level MAE of 1.8189 and RMSE of 3.8072. Its accuracy gains over the strongest sequence models were modest, but it produced no negative forecasts, while unconstrained learned models generated negative values in approximately 10.6% to 29.2% of raw forecast cells. The framework enforces nonnegative outputs and exact agreement between each model-predicted approach total and the sum of its component movements by construction, providing directly interpretable forecasts without clipping or coherence correction.