A Computational Predictive Model of Boko Haram
Abstract
A Computational Predictive Model of Boko HaramBoko Haram is perhaps one of the most deadly groups in West Africa. Though initially allied with Al-Qaeda, the group announced allegiance to ISIL in March 2015. It is estimated that Boko Haram~s reign of terror, predominantly in Northern Nigeria (where Islam is the dominant religion) has led to over 20,000 deaths and displaced over 2M civilians. We propose to develop the first ever data-driven, ensemble based predictive models that provide analysts with 1-4 months of advance warnings of the types of targets the Boko Harammight strike over the next few months. Our proposed research will lead to the following major advances:i) We will develop a general approach to ensemble based classification which mimics theidea of multiple classifiers working together collaboratively in order to reach aconsensual forecast. Such an ensemble based classification algorithm will be applicablenot only to our Boko Haram data, but to other data as well.ii) We will develop a general model to provide association rule based explanations forpredictions made by other predictors ~ these include not only our ECC predictor, butalso existing predictors that cannot generate good explanations for their forecasts. Theseexplanation techniques will be applicable not only to our Boko Haram data, but to otherdatasets as well.iii) We will develop the first ever detailed, data-driven, computationally-enabled predictivemodel of the behavior of Boko Haram. We will test and validate this model both via k~fold cross validation methods, as well as with live forecasts for a period of 6 months.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 23, 2016
- Source ID
- N000141612739
Entities
People
- V.s. Subrahmanian
Organizations
- Office of Naval Research
- United States Navy
- University of Maryland