New Mathematics Faculty Member Tracks Global Terrorism Patterns with NSF Grant

Study Begins with Focus on Nigeria and Afghanistan

For years, Mikyoung Jun has used her research in statistics for altruistic purposes, such as aiding atmospheric and climate science researchers with modeling to help them learn about air pollution and climate change to better understand our world.

Mikyoung Jun
Mikyoung Jun is developing statistical methods to model terrorism patterns across the world, beginning with Afghanistan and Nigeria.

Now, as a ConocoPhilips professor of data science at the University of Houston College of Natural Sciences and Mathematics, Jun is using her skills in one of her most challenging projects yet – developing statistical methods to model terrorism patterns across the world in order to provide scientists, policy makers and the general public useful information.

The National Science Foundation awarded Jun a $255,850 Algorithms for Threat Detection grant to do just this. With it, she will develop multivariate spatio-temporal Hawkes process models on a global scale to track terroristic activity.

Jun explains that traditionally, spatial statisticians assume the locations where data is measured are fixed.

“In general, this method that I’m developing, is a flexible framework for spatio-temporal point process models,” Jun said. “We are interested in developing statistical models for the location and time of random events. The point process models that I study describe and predict where and when events happen.”

The particular point process model she’s using is called the Hawkes process model. It has been commonly applied to earthquake modeling; however, she is adjusting the model to track terroristic events, both spatially and temporally.

Starting in Africa and the Middle East

“In the past, spatial statistics researchers have developed models more for local regions,” Jun said. “Because I’m developing a model on a global scale, the mathematical functions and associated statistical challenges are quite different.

Her team is first researching attacks in Nigeria and Afghanistan, partially because of their spatial inhomogeneity – meaning the terrorist attacks within the countries differ from one another.

“In Nigeria, there are two major terror groups: Boko Haram and Fulani extremists. Before we develop multivariate models, we are developing bivariate models to describe these two groups’ patterns jointly. Even spatially, the groups exhibit interesting repulsive patterns. If there is one area where one group is prominent, the other group is somewhere else. There is also spatial clustering of attack locations within each group.”

More Detailed Statistical Methods for Political Scientists

Jun became interested in using her statistical methods to model terrorism attacks, partially because she began to collaborate with Scott Cook, associate professor of political science at Texas A&M University.

The Global Terrorism Database, maintained by the University of Maryland since 1970, provides daily information on global terror attack locations, along with casualties, weapons used, responsible groups and many more details.

“They have a very comprehensive data set, but not many people have looked at the data with more rigorous statistical tools,” she said. “With the point process model, you can use the data at a much finer scale. For example, not only can you report 10 attacks in Nigeria on a given day, but you can pinpoint where an individual attack happened and whether there was a spatial and temporal triggering effect.”

Although she did not have a grand goal of making the world a better place going into the study, she hopes her method can be used by political scientists as a tool for the common good.

Her models, once they are made publicly available, could also be applied to a broad array of problems with similarly complex spatio-temporal patterns, such as crime, earthquakes and disease outbreaks.

- Rebeca Trejo, College of Natural Sciences and Mathematics