MI-ConGR-At: A Hybrid Spatio-temporal Deep Learning Approach for Traffic Data Classification

Traffic data classification is an important task for understanding and managing urban mobility patterns. It’s a critical component of Intelligent Transportation Systems. In this study, we have explored the application of deep learning models for classifying daily traffic data recorded at 963 sensing stations across 144 time intervals. The data recorded by sensors is spatiotemporal in nature and shows the road occupancy rate at a given point. The given task is to classify the input daily data instances into respective days of the week. We have proposed a hybrid approach combining mutual information classifier, convolutional networks, recurrent units, and attention mechanism. The results demonstrate that the proposed approach outperforms other baseline learning models explored in this study by a significant margin. The proposed model MI-ConGR-At has achieved a high classification accuracy of 99.09% and an F1 score of 0.99.