Abstract:
Terrorist attacks around the world are unpredictable and are a major threat for individuals as well as state. It creates violence, terror and damage to the general public and population causing unrest among individuals and society as well as social and economic failure in the region being attacked. It is still an ongoing threat for individuals and countries with much potential to grow due to more advanced, well planned and collaborated terrorist activities. The objective of such terrorist attacks is to cause fear, panic and terror among individuals and society. Counter terrorism are measures taken by the government and intelligence agencies to identify and predict the probability of an attack, in order to prevent an attack or identify the terrorist group involved in an attack. Counter terrorism also aims to identify and predict the possible areas of attack to minimize the chances and effects of terrorism. The measures taken to control and prevent terrorism can create social and economic stability. Security forces and intelligence agencies can take preventive measures before an attack and alert people to take safety measures to keep them away from such incidents or to avoid visiting the particular areas. Even when an attack happens this strategy can minimize the affects of terrorism. We proposed a method to forecast and predict the possible area of an attack in terms of city, country and region. Our proposed model is trained on deep learning model to predict the hotspots of terrorist attacks.We have used bidirectional recurrent neural network, bidirectional gated recurrent unit and bidirectional long short term memory network in order to identify the locations where terrorist attack can take place. The identification of location in terms of city, country and region provides a method for intelligence agencies to combat terrorism. Experimental results generated from bidirectional recurrent neural network, bidirectional gated recurrent unit and bidirectional long short term memory network shows that bidirectional long short term memory network outperforms other two models. We have evaluate the models using using different evaluation metrics such as accuracy, precision, recall and F1 score.
