Spatial intensity modelling of ISIS-related attacks using random forest regression: A GeoAI-based analysis of GTD data (2012–2019)

Spatial information is of crucial importance to understanding terrorist attacks and for security applications. Conventional risk assessments of terrorism are typically focused on retrospective identification of hotspots and may fail to fully account for complex spatial–temporal patterns. This research proposes an approach to modelling the relative intensity of ISIS/ISIL attacks across space using Random Forest regression, a type of Geospatial Artificial Intelligence (GeoAI). A subset of the Global Terrorism Database (GTD) from 2012–2019 is screened to develop a spatially aggregated data set with geographic coordinates, temporal information and some incident details. The modelling strategy estimates the intensity of events in a grid cell, with values normalized based on the count of previous events, allowing for an analysis of spatial patterns rather than the prediction of future events. Modelling results show that the Random Forest model captures spatial variability in historical attack intensity with a high level of fit (R² = 0.84; RMSLE = 0.213). Analysis of variable importance suggests spatial and temporal features play an important role, but this is contextualised by the spatial construction of the response variable. The research presents a repeatable GeoAI process for modelling spatial patterns of terrorism incidents and offers an interpretable model for exploratory spatial analysis and scenario-based spatial analysis development. The approach is not suitable for real-time prediction, but aids in understanding historical spatial concentration patterns in a complex threat environment.