Criticality assessment of terrorism related events at different time scales TENSOR clusTEriNg terroriSm actiOn pRediction

Law Enforcement Agencies (LEAs) are nowadays taking advantage of a wide range of information and intelligence sources (e.g., human intelligence (HUMINT), open source intelligence (OSINT), image analysis (IMINT)) to anticipate potential terroristic actions. Urban environments are nowadays associated with a wide range of vulnerabilities, which create fertile ground for terrorists planning actions against assets and/or citizens. These vulnerabilities stem from the characteristics of the urban environment (e.g., presence of civilians, availability of many and diverse physical infrastructures, complex social/cultural/governmental interactions, high value targets, etc.) have been repeatedly manifested as part of major terrorist attacks, which took place in some of the world’s most important cities (e.g., New York, London, and Madrid). The mitigation of security concerns in the urban environment is therefore a top priority in the social and political agendas of cities. ICT technologies provide help in this direction, for example through surveillance of urban areas, using the proliferating number of low-cost multi-purpose sensors in conjunction with emerging Big Data processing techniques for analyzing them. The thesis illustrates the TENSOR (clusTEriNg terroriSm actiOn pRediction) framework, a near real-time reasoning framework for early identification and prediction of potential threat situations (e.g. terrorist actions). The main objective of TENSOR is to show how patterns of strategic terroristic behaviors, identified analyzing large longitudinal data sets, can be linked to short term activity patterns identified analyzing feeds by “usual” surveillance technologies and that this fusion allows a better detection of terrorist threats. The framework consists of three different modules with the aim of collecting and processing information of the surrounding environment from a variety of sources including physical sensors (e.g. surveillance cameras) and “virtual” sensors (e.g. police officers, citizens). The proposed TENSOR framework processes information sources at different abstraction levels (e.g. sensor information, police inputs, external semantic crafted data sources) and, thru the proposed layered architecture, simulates the three main expert user roles (i.e. operational, tactical and strategic user roles), as indicated in the intelligence analysis domain literature. The framework transforms all the sensors gathered data into symbolic events of interest following a generic scenario-agnostic semantics for terrorist attacks described in literature as terrorist indicators. Thru different reasoning and fusion techniques, the framework proactively detects threats and depicts the situation in near real-time. The framework results have been tested and validated in the European project FP7 PROACTIVE.