Terrorism poses a significant global threat, with over 210,000 attacks since the late 1960s. The 2024 Global Terrorism Index indicates a concerning increase in both the frequency and lethality of attacks. While the impact of terrorism has varied over time, it remains a persistent challenge. Modern terrorism is characterized by decentralized actors, diverse targets (including cyber platforms), complex forms, and evolving tactics. Terrorist groups may be affiliated with religious extremism, separatist movements, or operate as lone wolves. Underlying factors such as religious extremism and socio-economic inequality fuel these activities. Current risk assessment and categorization methods are often subjective and lack comprehensive quantitative analysis. Existing quantitative studies on global terrorism often suffer from limitations such as short time spans, limited geographical scope, simplistic indicators, and subjective methodologies. This research aims to address these shortcomings by applying quantitative approaches to risk assessment and categorization of terrorist attacks.

The authors acknowledge existing international terrorism databases, including the GTD, ITERATE, TWEED, RDWTI, and WITS, but emphasize the GTD's comprehensiveness (210,454 incidents from 1970-2020) making it suitable for their quantitative analysis. They also acknowledge the GTD's limitations, such as selection bias stemming from reliance on media reports and the lack of contextual data on socio-political and economic factors influencing terrorist activities. Existing literature is reviewed, highlighting the predominantly qualitative nature of much terrorism research and the limitations of previous quantitative analyses, such as those relying on the analytic hierarchy process, fuzzy comprehensive evaluation, and K-mean clustering – methods criticized for subjectivity and lack of robust result evaluation. The authors aim to improve upon this existing research with their comprehensive and robust method.

The study utilizes the GTD dataset, containing 210,454 global terrorist incidents from 1970 to 2020. Data cleaning was performed to remove irrelevant or incomplete data, based on rules related to the 'doubtterr' variable, 'unknown' data entries, and the number of null records. An indicator system was constructed, selecting 22 variables related to terrorist attack risk (Table 1). Kendall's coefficient was used for correlation analysis among these indicators. Optimal weights for the indicators were determined using a combined subjective and objective weighting method based on moment estimation theory. Subjective weights were derived using the Analytic Hierarchy Process (AHP) and Order Relationship Analysis (ORA), while objective weights were calculated using the Entropy Weight (EW) and CRITIC methods. The optimal weights were obtained by minimizing the divergence between the integrated combination weight and the subjective and objective weights (Equations 2-9). Four risk assessment models were employed: Linear Weighted Evaluation (LWE), Fuzzy Comprehensive Evaluation (FCE), TOPSIS, and Particle Swarm Optimization Projection Pursuit Evaluation (PSO-PPE). The normalized comprehensive scores of these four models were averaged to provide a holistic risk assessment. Five clustering methods (FCM, CURE, DBSCAN, CLIQUE, and GMM) were used for risk categorization. The clustering results were evaluated using three internal criteria: Silhouette Coefficient (SC), Calinski-Harabaz Index (CHI), and Davies-Bouldin Index (DBI). Finally, kernel density estimation (KDE) was used for visual analysis of spatial risk patterns.

Data exploration revealed an upward trend in the number of terrorist attacks over the study period, with significant increases after 2000 (Figure 1). Explosives and firearms were the most common weapons, and bombing/explosion and armed assault were the most frequent attack types. Private citizens and property were the most common targets. Correlation analysis showed low correlation among the selected indicators. The optimal weights assigned high importance to casualties and consequences (38.12%), followed by incident location, perpetrator information, and target/victim information (Table 3). The top 10 riskiest attacks (Table 4) were identified, with the September 11th attacks ranking highest. Downward counterfactual events were also identified for each of these top 10 attacks (Table 5), revealing related plots or attempts that were thwarted. Clustering analysis identified four risk levels (Figure 4). Spatial analysis using KDE revealed four “turbulent cores” of high terrorist attack risk: Central Asia, Middle East & North Africa, South Asia, and Central America & Caribbean (Figure 5).

The findings demonstrate the utility of a quantitative, multi-method approach for risk assessment and categorization of terrorist attacks. The identification of the top 10 riskiest attacks and their downward counterfactuals provides valuable insights into patterns and trends in terrorism. The spatial distribution maps highlight regions requiring focused counter-terrorism efforts. The combination of multiple assessment models and clustering techniques provides a more robust and comprehensive evaluation than previous studies that relied on simpler, more subjective methods. The identified “turbulent cores” highlight geographic concentrations of terrorist activity that warrant targeted interventions.

This study provides a novel quantitative framework for assessing and categorizing terrorist attack risk. The identification of high-risk events and geographic hotspots offers crucial insights for counter-terrorism strategies. Limitations include the inherent biases and data limitations of the GTD. Future research could integrate diverse data sources, including socioeconomic indicators and social media data, to provide a richer understanding of the complex factors driving terrorism. Examining specific root causes and motivations for terrorism in these high-risk regions is essential.

The study's reliance on the GTD introduces limitations, including potential selection bias and underreporting, particularly in regions with limited media coverage or restricted information flow. The GTD also lacks contextual data on broader societal, political, and economic factors that can influence terrorism. The quantitative approach, while robust, does not fully capture the qualitative aspects of terrorism, such as the motivations and ideologies of terrorist groups. Furthermore, the study focuses on past data and may not fully capture the evolving dynamics of terrorism in real-time.