Spatial Distribution Assessment of Terrorist Attack Types Based on I-MLKNN Model

Abstract

Terrorist attacks are harmful to lives and property and seriously affect the stability of the international community and economic development. Exploring the regularity of terrorist attacks and building a model for assessing the risk of terrorist attacks (a kind of public safety risk, and it means the possibility of a terrorist attack) are of great significance to the security and stability of the international community and to global anti-terrorism. We propose a fusion of Inverse Distance Weighting (IDW) and a Multi-label k-Nearest Neighbor (I-MLKNN)-based assessment model for terrorist attacks, which is in a grid-scale and considers 17 factors of socio-economic and natural environments, and applied the I-MLKNN assessment model to assess the risk of terrorist attacks in Southeast Asia. The results show the I-MLKNN multi-label classification algorithm is proven to be an ideal tool for the assessment of the spatial distribution of terrorist attacks, and it can assess the risk of different types of terrorist attacks, thus revealing the law of distribution of different types of terrorist attacks. The terrorist attack risk assessment results indicate that Armed Attacks, Bombing/Explosions and Facility/Infrastructure Attacks in Southeast Asia are high-risk terrorist attack events, and the southernmost part of Thailand and the Philippines are high-risk terrorist attack areas for terrorism. We do not only provide a reference for incorporating spatial features in multi-label classification algorithms, but also provide a theoretical basis for decision-makers involved in terrorist attacks, which is meaningful to the implementation of the international counter-terrorism strategy.

AI-Driven Counter-Terrorism: Enhancing Global Security Through Advanced Predictive Analytics

“…Previous research in the fields of Machine Learning (ML) and Deep Learning (DL) has faced challenges in accurately predicting terrorist activities, particularly in the context of bilabel and multi-label classification tasks. This issue has been highlighted by [26] and discussed further by [27] in their respective studies. Convolutional Neural Networks (CNNs) have exhibited exceptional performance in tasks such as image classification, while other deep learning approaches have also demonstrated proficiency in multi-label classification.…”

Spatiotemporal multi‐graph convolutional network‐based provincial‐day‐level terrorism risk prediction

“…Uddin et al (2020) predicted essential features such as whether a terrorist attack will be successful and whether it will be a suicide attack based on deep neural networks. Zhao, Xie et al (2021) proposed a fusion of inverse distance weighting and multilabel k-nearest neighbor-based method to predict the grid-scale terrorist attack risk.…”

“…Zhao, Xie et al. (2021) proposed a fusion of inverse distance weighting and multilabel k ‐nearest neighbor–based method to predict the grid‐scale terrorist attack risk.…”

“…Loss‐sensitive learning is one of the crucial approaches to overcoming the challenge of imbalance, where the focal loss proposed by Lin et al. (2017) is one of the most known loss functions that has achieved promising results on a variety of imbalance classification tasks (Burhanudin et al., 2021; Zhao, Zhang, et al., 2021). Therefore, in this article, the focal loss is used to train the deep learning model, which is formulated as $$$$\begin{equation}{\mathrm{Focal\ loss}}\ \left( {v_i^t,p\ = \ f\left( {v_i^t} \right)} \right)\ = \left\{ { \def\eqcellsep{&}\begin{array}{@{}*{1}{c}@{}} { - \alpha {{\left( {1 - p} \right)}}^\gamma {\mathrm{log}}\left( p \right),\ {\mathrm{where}}\ v_i^t = 1}\\ { - \left( {1 - \alpha } \right){p}^\gamma {\mathrm{log}}\left( {1 - p} \right),\ {\mathrm{where}}\ v_i^t = 0} \end{array} } \right.\ ,\end{equation}$$$$where $$v_i^t$$ denotes the ground truth risk of province $$i$$ on day $$t$$.…”