BAST-MGCN: Integrating Background Attributes for Enhanced Spatio-temporal Prediction of Terrorist Attack Casualties

Terrorism seriously hinders the peace and stability of the international community. Predicting the potential casualties resulting from attacks by analyzing the spatiotemporal patterns and behavioral characteristics of terrorist attacks is a powerful safeguard for protecting the lives of innocent civilians. However, current spatiotemporal prediction methods for terrorist attacks focus on exploring the spatiotemporal dynamic correlations of terrorist attack data while ignoring the background attribute information of these events. Therefore, we propose a spatiotemporal multi-graph convolutional network called BAST-MGCN to predict casualties among innocent civilians caused by terrorist attacks. In our proposed method, we first use wavelet transform techniques to determine the main temporal cycles of terrorist attack spatiotemporal sequence data. Based on different time cycles, we construct spatial adjacency graphs of terrorist attack data and multiple trend graphs of background features. By constructing the BAST-MGCN multi-graph convolutional network, we extract the temporal correlations of different types of spatial features in terrorist attack spatiotemporal data. Experimental validation has confirmed the superiority of our method, showing that integrating background attribute information into spatiotemporal analysis of terrorist attack casualties can further enhance the predictive performance of the model.