An Experimental Study of Classification Algorithms for Terrorism Prediction

Abstract

Terrorist attacks are the biggest challenging problem for the mankind across the world, which need the wholly attention of the researchers, practitioners to cope up deliberately. To predict the terrorist group which is responsible of attacks and activities using historical data is a complicated task due to the lake of detailed terrorist data. This research based on predicting terrorist groups responsible of attacks in Egypt from year 1970 up to 2013 by using data mining classification technique to compare five base classifiers namely; Naïve Bayes (NB), K-Nearest Neighbour (KNN), Tree Induction (C4.5), Iterative Dichotomiser (ID3), and Support Vector Machine (SVM) depend on real data represented by Global terrorism Database (GTD) from National Consortium for the study of terrorism and Responses of Terrorism (START). The goal of this research is to present two different approaches to handle the missing data as well as provide a detailed comparative study of the used classification algorithms and evaluate the obtained results via two different test options. Experiments are performed on real-life data with the help of WEKA and the final evaluation and conclusion based on four performance measures which showed that SVM, is more accurate than NB and KNN in mode imputation approach, ID3 has the lowest classification accuracy although it performs well in other measures, and in Litwise deletion approach; KNN outperformed the other classifiers in its accuracy, but the overall performance of SVM is acceptable than other classifiers