A hybrid deep learning-based framework for future terrorist activities modeling and prediction

Abstract

Terrorism has led to massive humanitarian and economic crisis due to dire events that affected many countries and caused thousands of deaths and critical damages. In literature, various Artificial Intelligence (AI) based research works have been proposed and enhanced to counter and predict terrorist activities. Practically, Machine Learning (ML) techniques are the most applied. However, with the increasing of the complexity and volume of data, ML algorithms fail to detect and predict accurately terrorist activities. Thus, for understanding the behavior of terrorist actors, we proposed a hybrid Deep Learning (DL) platform based on Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) models to learn the temporal features from the Global Terrorism Database (GTD) and to predict future terrorist activities characteristics. The GTD is a well-known database which contains around 190,000 terrorist events and incidents around the world since 1970 until 2020 and incorporates multiple factors, such as the type of weapons used, the attack is successful or not, the kind of attack, and the category of terrorist. First the CNN is used extract complex features of the data, and then these features are forwarded to LSTM model to learn the temporal relationship of the data. Simulation results show that the CNN-LSTM model achieves superior performance for bi-classification tasks which achieves an accuracy more than 96%, while the DNN outperforms the hybrid aforementioned model with accuracy of 99,2% the multi-classification task of predicting terrorist activities. The proposed model shows also a correlation between the occurrence of attacks with the type of weapons used and can accurately predict the type of terrorist attacks with their success rate.