Applying Hybrid Lstm-Gru Model Based on Heterogeneous Data Sources for Traffic Speed Prediction in Urban Areas

Area of Science:

• Computer Science
• Artificial Intelligence
• Urban Planning

Background:

• The Internet of Things (IoT) enables vast data generation in smart cities via diverse sensors.
• Integrating heterogeneous data from various administrative domains and APIs is crucial for Intelligent Transport Systems (ITS).
• A hybrid feature space is essential for effective traffic prediction in complex urban environments.

Purpose of the Study:

• To develop a comprehensive algorithm for integrating heterogeneous data sources into a hybrid spatial-temporal feature space.
• To comparatively analyze the performance of various deep learning models for time series geospatial traffic prediction.
• To identify the most effective deep learning architecture for smart city traffic forecasting.

Main Methods:

• Developed a novel algorithm for integrating sensor, service, and exogenous data into a unified spatial-temporal feature space.
• Conducted exploratory data analysis on urban sensor data.
• Applied and compared Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Convolutional Neural Network (CNN), and hybrid deep learning models.

Main Results:

• The hybrid LSTM-GRU model demonstrated superior performance in traffic prediction.
• Achieved a Root Mean Squared Error (RMSE) of 4.5 and a Mean Absolute Percentage Error (MAPE) of 6.67% with the hybrid LSTM-GRU model.
• Comparative analysis highlighted the effectiveness of hybrid deep learning architectures for geospatial time series data.

Conclusions:

• The proposed data integration algorithm effectively creates a hybrid feature space for smart city applications.
• Hybrid deep learning models, particularly LSTM-GRU, offer significant improvements in traffic prediction accuracy.
• This research contributes to the advancement of Intelligent Transport Systems (ITS) through enhanced data integration and predictive modeling.