Forecasting the risk of terrorist attacks based on machine learning algorithms

Abstract

This article is devoted to the analysis and prediction of the risk of terrorist acts based on a comparison of various machine learning algorithms. In order to determine the most important indicators, more than thirty external and internal risk factors are comprehensively considered by quantifying them and an initial set of initial data is built. The study analyzes multidimensional socio-economic and political data for 136 countries for the period from 1992 to 2020. Four indicators are also predicted, reflecting the expected success of terrorist attacks, the likelihood of socio-economic consequences and general damage from terrorism. In addition to the classical analysis models, the effectiveness of the other four machine learning algorithms that can be used to analyze multidimensional data is compared. To predict the risk of terrorist attacks, a random forest model is created, and the effectiveness and accuracy of the model are evaluated based on statistical criteria. To determine the most important initial indicators, the method of recursive elimination of features in a random forest was used. The main result of this study is to identify the most important indicators for predicting the risk of terrorism and to reduce redundant indicators, which makes it possible to improve understanding of the main characteristics of attacks. Meanwhile, the results show that it is necessary to take appropriate proactive measures not only in the form of forceful detention, intelligence and response operations, but also to improve the stability of the state, achieve social balance and improve the quality of life of citizens.