Terrorist organization behavior prediction algorithm based on context subspace

Abstract

Researchers have developed methods to predict a terrorist organization's probable actions (such as bombings or kidnappings). The terrorist organization's actions can be affected by the context of the organization. Thus, the organization's context variables can be used to improve the accuracy of forecasting the terrorist behavior. Those algorithms based on context similarity suffer a serious drawback that it can result in the algorithm's fluctuation and reduce the prediction accuracy if not all the attributes are detected. A prediction algorithm PBCS (Prediction Based on Context Subspace) based on context subspace is proposed in this paper. The proposed algorithm first extracts the context subspace according to the association between the context attributes and the behavior attributes. Then, it predicts the terrorist behavior based on the extracted context subspace. The proposed algorithm uses the improved spectral clustering method to obtain the context subspace. It concerns the distribution of the data samples, the label information and the local similarity of the data in the process of extracting the subspace. Experimental results on the artificial dataset and the MAROB dataset show that the prediction method proposed in this paper can not only improve the prediction accuracy but also reduce the prediction fluctuation. © 2011 Springer-Verlag.