Upcoming talk

On my upcoming talk at QMDNS (Linky)...

Title: Modeling and Detection of Sudden Spurts in the Activity Profile of Terrorist Groups

Abstract: The main focus of this work is to detect sudden spurts in the activity profile of a terrorist group and track it over a period of time. Towards this goal, a $d$-state hidden Markov model (HMM) that captures the strength of the group and thus its activity profile is developed. The simplest setting of $d = 2$ corresponds to the case where the strength is coarsely quantized as Active and Inactive, respectively. Two strategies for spurt detection and tracking are developed here: a model-independent strategy that uses the exponential weighted moving-average (EWMA) filter to track the strength of the group as measured by the number of attacks perpetrated by it, and a state estimation strategy that exploits the underlying HMM structure. The EWMA strategy is robust to modeling uncertainties and errors, and tracks persistent changes (changes that last for a sufficiently long duration) in the strength of the group. On the other hand, the state estimation strategy tracks even non-persistent changes that last only for a short duration at the cost of learning the underlying model. Case studies with real terrorism data from open-source databases are provided to illustrate the performance of the two strategies.

Short Summary: Existing work on activity profile modeling fall under three categories: i) classical time-series techniques of fitting trend, seasonality and stationary components, ii) a technique that goes by the name group-based trajectory analysis (derived primarily from statistical methods in sociology research) where terrorism trends with similar developmental paths are identified, and iii) self-exciting point process models. The first method has been around from the early 80s in different forms, the second method corresponds to early 2000s when many open-source terrorism databases came into the picture, while the third method is rather recent. Of these, the third method is the most exciting (pardon the pun!) as it allows the full-blown hammer of spatio-temporal point process techniques combined with statistical methodologies in hypothesis testing, inferencing, parametric and non-parametric methods, etc. In some sense, this modeling method has been motivated by similar trends in gang warfare, seismology, epidemiology, etc., where self-exciting models have really pushed the frontiers of statistical inferencing. In terms of negatives, the model has too many parameters that have to be learned for all these inferencing techniques to work nicely.

My work is motivated by two central issues. One, terrorism (despite the huge media attention) is a rare phenomenon with data-collection a painful manual exercise (despite the rise of many open-source databases). Thus, parsimonious models driven by simplistic hypotheses make really good sense in this business. Two, while self-exciting models make perfect sense for gang warfare where revenge attacks can be nicely modeled as self-excited and in seismology where aftershocks are driven by primary shocks, it is not clear if this is the best framework to be used for terrorism modeling. In this context, I propose an alternate HMM framework where the hidden states model certain qualitative attributes of the terrorist group which could undergo a shift over time and thus drive the activity profile in different ways. I show via statistical techniques that this model is a good competitor for self-exciting models and has the added advantages of being described by few parameters (four, in the simplest non-trivial setting), and being motivated by a simpler set of hypotheses on the bare basics of how terrorism works. To rephrase Peter Diggle, at the end of the day, the objectives of a statistical analysis should be determined by the objectives in collecting a certain dataset under question. To address this issue, I show that the proposed approach detects and tracks spurts in the activity profile that can be easily identified with certain geopolitical realities. This is done with data on FARC, Shining Path and the Indian Maoists.