The unpredictable nature of terrorist attacks has gathered significant attention from governments, security experts, counterterrorism specialists, and international organizations such as NATO, owing to their devastating impact on human life and societal stability. This thesis advances the use of machine learning (ML) and explainable artificial intelligence (XAI) in terrorism prediction, addressing the challenge of balancing high predictive performance with interpretability in security‑critical environments. While prior research has applied ML to terrorism data, limited work has produced transparent or quantifiable explanations capable of building user trust. This study addresses that gap by applying supervised ML models to three core tasks using the Global Terrorism Database: terrorist group classification, suicide attack prediction, and target‑type prediction. Benchmark models were evaluated alongside first‑time applications of advanced algorithms, with model outputs interpreted through permutation feature importance, Shapley Additive Explanations, and Anchors. A novel contribution of this thesis is the quantification of interpretability using Anchors, measuring precision, coverage, and rule complexity to assess explanation reliability and operational usefulness. Results show that ensemble‑based models such as Random Forest, LightGBM, and Histogram‑based Gradient Boosting achieved strong predictive performance, with ROC‑AUC scores exceeding 0.90 in several cases. However, performance metrics alone proved insufficient. Explainability techniques revealed consistent model reliance on spatio‑temporal features, attack characteristics, target infrastructure, and fatalities, providing actionable insights for counterterrorism practitioners. Anchors and SHAP force plots generated local, instance‑level explanations, while quantitative evaluation highlighted trade‑offs between rule precision and coverage, underscoring the need to balance specificity with generalizability. Overall, this research demonstrates that interpretable ML can enhance operational decision‑making by explaining black‑box predictions in terrorism contexts. The findings provide transparent, evidence‑based insights to support intelligence officers and policymakers. Limitations include reliance on historical and post‑event GTD data, while future work should incorporate real‑time sources, alternative XAI methods, and human‑in‑the‑loop decision‑support systems.
A Systematic Terrorism Prediction Framework Using Explainable Artificial Intelligence
Lamptey, O. (Author). 5 Jun 2026
Student thesis: Doctoral Thesis