Hourly Origin-Destination Matrix Estimation Using Intelligent Transportation Systems Data and Deep Learning

Area of Science:

• Transportation Science
• Urban Planning
• Data Science

Background:

• Traditional origin-destination (OD) demand surveys are costly and time-consuming.
• Intelligent Transportation Systems (ITS) generate diverse data suitable for demand prediction.
• Leveraging big data from ITS offers significant benefits for transportation planning.

Purpose of the Study:

• To predict the OD matrix for Tehran metropolis using various ITS data sources.
• To evaluate the performance of different machine learning (ML) models for OD matrix prediction.
• To identify the most accurate ML model for OD matrix estimation.

Main Methods:

• Utilized data from automatic number plate recognition (ANPR) cameras, smart fare cards, loop detectors, and GPS navigation software.
• Incorporated socio-economic, demographic, and land-use characteristics of zones.
• Developed and compared five ML models, including convolutional neural networks (CNN).

Main Results:

• The CNN model demonstrated the highest accuracy in predicting the OD matrix.
• CNN achieved the lowest Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE).
• The CNN-predicted OD matrix exhibited the highest structural similarity to the ground truth matrix.

Conclusions:

• Machine learning, particularly CNN, is effective for predicting urban OD matrices using ITS data.
• CNN offers a superior approach for accurate and reliable travel demand forecasting.
• The findings support the integration of diverse ITS data for enhanced transportation planning.