A Bayesian network analysis of aviation terrorism attack risks

Abstract

In recent years, statistical and mathematical models, e.g. machine learning, have been used in counterterrorism medicine research in order to understand the characteristics of terrorist incidents. The objective of this study was to assess the main risk factors related to the number of injuries using a Bayesian network analysis. Data on 338 aviation terrorism incidents between the year 2000 and 2020 were collected from the Global Terrorism Database. Seven aviation sector-specific security-affecting factors (SRIFs) were analyzed: country, region, attack type, location, property damage, injuries, and fatalities. A tree-augmented na & iuml;ve Bayes network analysis was used to define the association among the seven SRIFs with the number of injuries as training node. "Country" and "fatality" exert the greatest influence on the "injured" node, each accounting for more than 24% of the entropy reduction. This suggests that national-level factors and the severity of fatalities are key determinants in predicting injury counts. "Property damage" also demonstrated a substantial effect, contributing over 20% to the overall reduction in uncertainty. "Attack type," "region," "weapon," and "location" had comparatively lower mutual information values, indicating a weaker, but still notable, influence on injury outcomes. These findings highlight the heterogeneous contributions of SRIFs to injury prediction. Bayesian network analysis offers valuable insight into the complex interdependencies among aviation terrorism risk factors. The findings highlight the heterogeneous contributions of different SRIFs to injury prediction. These results can inform practitioners, researchers, and policymakers by supporting more proactive, evidence-based strategies for aviation security and emergency preparedness.