Inferring origin-destination distribution of agent transfer in a complex network using deep gated recurrent units

Area of Science:

• Complex Systems Analysis
• Network Science
• Machine Learning

Background:

• Predicting agent origin-destination (OD) probability distributions is crucial for managing complex systems.
• Existing statistical estimators for OD prediction often suffer from underdetermination, lacking a generalizable approach.
• Current methods for improving OD prediction accuracy are limited in scope and generalizability.

Purpose of the Study:

• To introduce a novel deep neural network framework with gated recurrent units (DNNGRU) for network-free OD prediction.
• To investigate the impact of network topologies on OD prediction accuracy using the proposed DNNGRU.
• To demonstrate the superior performance of DNNGRU compared to existing methods and alternative neural network architectures.

Main Methods:

• Developed a deep neural network framework with gated recurrent units (DNNGRU).
• Trained the DNNGRU using supervised learning on time-series data of agent volumes passing through network edges.
• Evaluated DNNGRU performance by analyzing its accuracy in predicting OD probability distributions across various network topologies and data scenarios.

Main Results:

• The DNNGRU framework achieves high accuracy in predicting origin-destination (OD) probability distributions.
• Prediction accuracy is influenced by the degree of path overlap between different OD pairs within the network.
• The proposed DNNGRU consistently outperforms existing OD prediction methods and alternative neural network architectures.

Conclusions:

• The proposed DNNGRU offers a general and effective approach for OD prediction in complex systems.
• The network-free DNNGRU framework provides near-optimal performance, surpassing current state-of-the-art methods.
• This study highlights the potential of deep learning for advancing OD prediction accuracy and understanding network dynamics.