Machine learning-based target prediction for terrorist attacks

Abstract

To address the problem that terrorist attacks are frequent and attack targets are widely difficult to predict, we use multidimensional features of terrorist attacks to predict terrorist targets based on a quantitative statistical analysis of data in the global terrorism database (GTD) from 1970 to 2019. In this paper, a machine learning-based prediction model (i.e., a classifier framework) is proposed. The model pre-processes GTD data and uses the OneHotEncoder and KBinsDiscretizer methods for data category transformation, classifies terrorist targets by four algorithms—SVM, decision trees, KNN and DNN. Each algorithm was trained with their parameters optimised, and the performance of the proposed model was tested and evaluated. The experimental results show that the prediction model achieved good performance in predicting the attack targets. The neural network had the best prediction performance in predicting twenty-two attack targets with 100% accuracy, indicating that the proposed model is accurate and effective.