Optimized CapsNet for Traffic Jam Speed Prediction Using Mobile Sensor Data under Urban Swarming Transportation

Area of Science:

• Intelligent Transportation Systems (ITS)
• Machine Learning
• Urban Planning

Background:

• Urban swarming transportation (UST) involves mixed traffic (vehicles, pedestrians), complicating traffic jam prediction.
• Existing intelligent traffic system (ITS) research primarily focuses on single road networks, not complex UST environments.
• Accurate traffic jam prediction is crucial for efficient urban mobility and traffic management.

Purpose of the Study:

• To propose a novel road network wide (RNW) traffic prediction method for urban swarming transportation (UST).
• To develop an optimized deep learning model for analyzing spatial-temporal traffic data.
• To evaluate the proposed method's effectiveness using real-world urban traffic data.

Main Methods:

• Utilized citizens' mobile GPS sensor records for RNW traffic prediction.
• Developed a data preprocessing technique to convert traffic data into spatial-temporal images.
• Proposed OCapsNet, a revised capsule network (CapsNet) with enhanced convolution layers and dynamic routing, for traffic jam speed prediction.

Main Results:

• OCapsNet demonstrated superior performance compared to Convolutional Neural Network (CNN) and the original CapsNet.
• The proposed method achieved higher accuracy and precision in predicting traffic jam speeds.
• Experiments were validated using real-world urban traffic data from Jakarta.

Conclusions:

• The OCapsNet model offers a significant advancement in predicting traffic jam speeds within complex urban swarming transportation (UST) scenarios.
• RNW traffic prediction using optimized deep learning models is effective for managing urban traffic.
• The findings provide a valuable contribution to the field of intelligent transportation systems (ITS).