Tensor Decomposition for Spatial-Temporal Traffic Flow Prediction with Sparse Data.

Area of Science:

• Transportation Engineering
• Data Science
• Urban Planning

Background:

• Urban transport surveillance is crucial for traffic control and travel planning.
• Traffic flow prediction is essential for optimizing urban mobility.
• Data sparsity presents a significant challenge in accurate traffic flow prediction.

Purpose of the Study:

• To develop an effective traffic flow prediction method that overcomes data sparsity.
• To leverage spatial and temporal correlations in transportation traffic data.
• To improve the accuracy and robustness of traffic surveillance systems.

Main Methods:

• Modeling traffic flow using a fourth-order tensor incorporating location, time of day, day of week, and week of month.
• Estimating correlations across tensor dimensions.
• Utilizing a gradient descent-based algorithm for prediction, addressing spatial and temporal data gaps.

Main Results:

• The proposed method demonstrates superior prediction accuracy compared to baseline approaches.
• Prediction accuracy shows minimal degradation with increasing percentages of missing data.
• The method effectively handles missing data in neighboring roads and across multiple days.

Conclusions:

• The developed traffic prediction method is robust and effective for sparse data scenarios.
• This approach enhances the reliability of transportation traffic surveillance systems.
• The findings support the application of this method in real-world urban traffic management.