Spatiotemporal multi-graph convolutional network-based provincial-day-level terrorism risk prediction.

Area of Science:

• Computational Social Science
• Network Science
• Artificial Intelligence

Background:

• Terrorism risk prediction is vital for effective counter-terrorism strategies.
• Terrorism risk exhibits complex spatiotemporal contagious characteristics influenced by interprovincial attacks and internal/external factors.
• Existing models struggle to capture the multidimensional, non-Euclidean relationships inherent in terrorism diffusion.

Purpose of the Study:

• To propose a novel spatiotemporal graph convolutional network (STGCN)-based extension method for predicting daily terrorism risks.
• To model the complex contagious diffusion of terrorism risk across provinces.
• To enhance the accuracy of terrorism risk forecasting by incorporating multidimensional spatiotemporal correlations.

Main Methods:

• Developed an extended spatiotemporal graph convolutional network (STGCN) incorporating long short-term memory and self-attention layers for temporal dynamics.
• Constructed three graph structures (distance, root cause similarity, self-excited) to represent interprovincial contagious processes.
• Utilized one-dimensional convolutional neural network kernels and spectral graph convolution modules to capture temporal and spatial features, respectively.

Main Results:

• The proposed extended STGCN method demonstrated superior effectiveness in predicting terrorism risk compared to other machine learning models.
• Experimental results on Afghanistan terrorist attack data (2005-2020) validated the model's predictive capabilities.
• The study highlighted the critical importance of capturing comprehensive spatiotemporal correlations among provinces for accurate risk forecasting.

Conclusions:

• The extended STGCN provides a powerful tool for understanding and predicting terrorism risk diffusion.
• Findings offer valuable insights for counter-terrorism management, emphasizing both long-term root cause mitigation and short-term situational prevention.
• Accurate terrorism risk forecasting necessitates a holistic approach considering interconnected spatiotemporal factors.