Terrorist behaviour profiling: a machine learning approach

This paper answers the question of whether it is possible through advanced machine learning techniques to deduce a risk profile for possible terrorists. Starting from the data on a particular individual, if whether or not they are part of a terrorist organization, the aim of the work is to extract patterns significant within the data, in order to study a subject, or a group of subjects, by building a profile of risk. An Ensemble-type algorithm such as Extreme Gradient Boosting (XGBoost) is applied to a data set of 950 subjects under investigation or indictment for crimes related to terrorism: bomb attacks, explosive material, murder and so on. The features concerning the subjects include their personal data, race, sex, city where the crime was perpetrated and other details. Starting from this information, the goal is to draw a description, in order to identify future attackers or events. Although the task is very difficult and the problem highly complex, good results have been achieved that bode well in future applications of methods of this type, always considering that as more data is available to build an analytical model, the greater the likelihood of the predictive effectiveness of the model.