Interpretable machine learning-based terrorist attack success rate prediction

The success of terrorist attacks reflects the capability of terrorists and the vulnerability of the security defense, explainable prediction of the average attack success rate at the country-annual level is crucial for governments. In this study, terrorist attack data from 146 countries between 2002 to 2020 was obtained from the global terrorism database (GTD), and a two-stage prediction task was conducted. First, multiple machine learning models, including XGBoost and Random Forest (RF), are used to predict the average success rate of terrorist attacks in the next year, considering terrorism root factors and statistical results from the previous year. The results show that the RF model performs the best. Second, the prediction outputs of the RF model are explained using interpretable methods, including Accumulate Local Effect (ALE) and SHapley Additive exPlanation (SHAP), to provide counterterrorism insights applicable to countries around the world.