Terrorist Group Classification using Logistic Regression

Abstract

Terrorism is a global challenge and the responsible terrorist groups has become global enemies. The world most challenging threat nowadays are these deadly terrorist attacks which carried out by these well-organized terrorist groups across the world. Only few nations can claim not to have been affected by terrorism. It is a complicated task to accurately classify and predict the accountable terrorist group due to unknown terrorist groups in global terrorism historical data. Terrorist attacks are increasing day by day with a great pace around the world and these attacks are one of the severe problems that humanity is facing in the current scenario. Nowadays terrorists are well organized into well-known groups and these groups are becoming more global, dynamic and networked. Different terrorist groups have different attack behaviors and for counter terrorism to be affected we should classify these terrorist groups according to their attacks. Fortunately, in the developed countries there is a high percentage of attacks being reported and the opposite for developing countries where mostly the important information is missing or incomplete, but the data of these attacks is maintained globally in a database. This research focuses on the terrorist group classification of unknown terrorist groups from Global Terrorism Database (GTD) using state of the art machine learning algorithms, specifically text-based feature generation using TF-IDF and multi-class logistic-regression these are widely used for classification due to their high accuracy and maturity. Additionally, Descriptive Statistical Analyses is performed before and after classification to get complete picture of global terrorism and to capture macrotrends. Therefore, with TF-IDF and logistic regression we have trained and evaluated our proposed model by considering attack summaries written by an analyst against each attack and other most relevant features to accurately predict the associated terrorist group. After the prediction of unknown terrorist groups Descriptive Statistical Analyses has done again to see the effectiveness and increase in the information value in the dataset. Performance evaluation of the proposed model as well as comparison of existing state-of-the-art machine learning models is also done. Our work focuses on text-based classification solution to accurately fill in the missing information for an attack where it would not have been known otherwise. Our main objective is to increase the model’s ability to correctly identify the terrorist group based on the attack summaries an analyst has written, this work will greatly increase the information value of the Global Terrorism Database (GTD).