No Batteries Required - Home
Fighting Terror with Predictive Analytics
Ray Kahn
APR 26, 2013 12:23 PM
A+ A A-

Predictive Analytics

Could we have prevented the Boston Marathon terrorist attacks which left three people dead and scores of others severely injured?

As I try to recover from an acute back injury I continue to ponder this question: Could predictive analytics have helped event organizers and local and Federal law enforcement agencies prepare for and prevent this act? After all this was not a “Black Swan” event – a hard-to-predict and rare event - we have witnessed and experienced many terrorist attacks against the homeland and our interests for many years now. Frankly as I read more about the emerging profiles of the suspects, specifically that of the mastermind of the attacks, plus the attention that such events as Boston Marathon elicit I don’t see why not.

What Is Predictive Analytics

Predictive analytics is the process of studying past events to predict the likelihood of future events. It uses statistical analysis to extract information from data to discover or predict trends and patterns (of course this is an oversimplification of the discipline).

The building blocks of predictive analytics are predictive models which combine data and mathematics to create a mapping function between a set of input data variables and a response variable.  The mapping function is a predictive modeling technique - back-propagation neural networks, support vector machines or decision trees – where data is used repeatedly to train the model to learn the mapping between the input variables and desired output (tweaking the model).

So imagine that we want to create a predictive model that is able to tell the likelihood of the next terrorist attack. Such a model needs to be generic, be accurate and work of course, which means that our model would need plenty of good quality data to be fully validated. Such data can be compiled from different regions of the world that have high incident rates of terrorism: Israel, Afghanistan and Iraq are just a few countries that suffer disproportionately from terrorist acts. A terrorist act is one that uses violence as a tactic to achieve a psychological effect. So for the purposes of our model the act of terror is the response variable. Our model would not need to predict who may commit such an act – would that be a more challenging problem? - but rather the likelihood of such an act. To build our model we also need a dataset describing the event. The dataset could contain information about the scale of the event (local, regional, national or international), the importance of the event (Olympic vs. a high school’s track and field event), visibility of the event (is it being broadcast locally, nationally or internationally? Is there a lot of interest in the event?), utility of the terror act to the terrorist(s) during the event, location of the event (urban vs. rural), degree of law enforcement and intelligence resources dedicated for the security of the event, local population diversity (different ethnicity/religious affiliations- whether we like it or not people are the drivers of these horrific acts), local minority/migrant unemployment levels, etc. Of course coming up with a complete list would be impossible because of unknown unknowns, and thus the need for iterative improvement of all predictive models.

The ultimate goal of predictive analytics is to prevent an event or outcome from happening in the first place. A predictive model will have false positives and negatives regardless of how much “training” the model has gone through and this leads me to conclude that any model we come up with would ultimately require human interpretation, which is a good thing. Expert knowledge plus data driven knowledge should make for a better model.


Predictive analytics has been widely used in financial and insurance industries as well as in CRM systems to prevent customer churn. It has been a potent tool in fighting credit card fraud as well as helping insurance companies devise accurately priced insurance policies and catch fraudulent claims. And more and more companies today use this tool in their product R&D, recommendation services and targeted marketing campaigns. Predictive analytics has had its share of failures such as failing to predict and prevent the 2008 financial crisis. But it is still a very powerful tool that can help our law enforcement agencies fight terrorism.


Alex Guazelli. 2012. Predicting the future, Part 1: What is predictive analytics?

Alex Guazelli. 2012. Predicting the future, Part 2: Predictive modeling techniques.


Ransom Weaver, Barry G. Silverman, PhD, Hogeun Shin, Rick Dubois. Modeling and Simulating Terrorist Decision-making: A “Performance Moderator Function” Approach to Generating Virtual Opponents.


[%= name %]
[%= createDate %]
[%= comment %]
Share this:
Please login to enter a comment: