Identifying high-risk areas of terrorism in the United Kingdom and France through machine learning

Abstract

In recent years, the increasing threat of terrorism in Western Europe has led to various counter-terrorism measures being introduced to reduce the level of risk posed. To aid these efforts and help better allocate the resources being implemented, academics have undertaken research to attempt to understand the spatial dynamics of terrorism and to predict areas where attacks are likely to occur. Previously, various analytical techniques have been applied at a fine-scale to identify these locations. However, despite yielding accurate results in other fields, a prediction technique that has yet to be fully utilised in the fine-scale spatial analysis of terrorism is Machine Learning. Consequently, this study aimed to use Machine Learning to identify high-risk areas of terrorism at a fine-scale in the United Kingdom and France. To achieve this aim, historic data from 1996 to 2016 was used to train and test the performance of a common Machine Learning algorithm, ‘Random Forest’. The application of this algorithm was successful, as high levels of accuracy (97.9%, 96.8% and 95.7%) and specificity (90%, 88%, 87%) were achieved when predicting high-risk areas. Furthermore, when compared to a conventional method for prediction (logistic regression), Machine Learning was found to be ~9% more accurate, and 2% more specific, at predicting high-risk locations. Therefore, the results of this study show promise for utilising Machine Learning in future terrorism-related spatial research and ultimately, have shown that applying such techniques could potentially contribute towards the better allocation of counter-terrorism resources in the UK and France.