Mining the Global Terrorism Dataset using Machine Learning Algorithms

From the beginning of the present century, terrorist attacks have represented a massive concern for both developed and developing countries alike. Therefore, countries spare no effort to use all available means to fight and eradicate them. Due to this, we aim through this paper, to take the advantages of the available data mining and machine learning techniques, by applying them to the Global Terrorism Database (GTD) dataset in order to acquire valuable information about the predicted attacks and attackers. We believe that these models will be of great benefit when utilized by governments and intelligence agencies; since they help them to make pre-emptive strikes against terrorist groups quickly and over a short period of time. In our paper, we focus on two tasks: predicting the success of an attack and predicting the identity of the terrorist organization that is behind the attack. To implement this, we adopted three machine learning algorithms: K-nearest neighbor (KNN), Naive Bayes (NB) and Random Forest (RF), which we used to train our models. More specifically, each algorithm was used to construct two models, one for each task, where the data were sampled for the first model using the holdout method, while for the second model cross-validation was employed. In the end, we compared the performance of the models in terms of accuracy, precision, recall and F-measure metrics. We noticed that the RF models outperformed the other models, while the NB models were the least efficient among the three algorithm models.