Regionally optimized global terrorism prediction with climate and spatiotemporal information: a grid-month framework

Terrorist attacks continue to pose significant threats to national public safety and global peace. Accurately predicting terrorist activity at a fine geographic scale is essential to implement effective measures and allocate sufficient resources for risk mitigation. Here, we developed a novel grid-month spatiotemporal framework to accurately predict global terrorism, by utilizing 21 years of global thematic data that include 31 distinct terrorism-related risk factors, climate factors, and spatiotemporal statistical data on terrorism. Specifically, given the marked regional heterogeneity worldwide, we employed automated machine learning (AutoML) models to independently identify the optimal model and hyperparameter configuration for each geographic region. We further applied Accumulated Local Effects to explore the potential relationships between climate factors and terrorism. Our findings demonstrate that integrating AutoML, climatic variables, and spatiotemporal terrorism features improved predictive performance across global regions by an average of approximately 20%, with regional AUROC values ranging from 0.88 to 0.96. Additionally, extreme temperatures, reduced precipitation, and frequent disasters are strongly associated with an increased risk of global terrorism. The impact of these climate factors on terrorist attack risks varies significantly across regions. These insights assist policymakers in formulating effective counter-terrorism strategies based on the heterogeneous influences of risk factors in different regions.