A multi-modal geospatial-temporal LSTM based deep learning framework for predictive modeling of urban mobility...

Area of Science:

• Urban planning and transportation science
• Artificial intelligence and machine learning
• Geospatial data analysis

Background:

• Accurate urban mobility prediction is essential for efficient resource allocation and urban development.
• Existing methods often struggle to capture complex spatiotemporal dynamics in urban environments.

Purpose of the Study:

• To propose a novel framework, GeoTemporal LSTM (GT-LSTM), for enhanced urban mobility prediction.
• To effectively integrate temporal dependencies with geographic information for improved forecasting.

Main Methods:

• Developed a multi-modal framework combining attention mechanisms and Recurrent Neural Networks (RNNs).
• Utilized attention mechanisms to dynamically weight geographic features.
• Employed LSTM layers to model sequential time-series data patterns.

Main Results:

• Achieved a 15% reduction in Mean Absolute Percentage Error (MAPE).
• Demonstrated a 20% reduction in Root Mean Square Error (RMSE) compared to traditional methods.
• Outperformed existing techniques like Convolutional LSTM and Graph Convolutional Networks.

Conclusions:

• GT-LSTM effectively captures complex spatial and temporal dynamics in urban mobility.
• The framework offers significant potential for real-time prediction and improved urban planning.
• Provides valuable insights for policymakers and transportation authorities to enhance system efficiency.