A quantitative geospatial analysis of the risk that Boko Haram will target a school.

Abstract

We provide a novel quantitative geospatial analysis of school attacks perpetrated by Boko Haram in Nigeria. Such attacks are used by Boko Haram to kidnap boys (for potential use as child soldiers and suicide bombers) and girls (for potential use as domestic servants, as sex slaves, and suicide bombers). We first build a novel geospatially tagged data set spanning almost 15 years (July 2009 to April 2023) of data not only on Boko Haram attacks on schools (our dependent variable) but also a set of 15 independent variables (or features) about other attacks by Boko Haram, locations of security installations, as well as socioeconomic and geospatial characteristics of the regions around these schools. Second, we develop a univariate statistical analysis of this data, showing strong links between three broad factors affecting attacks on schools: Security presence in and around a school, the Boko Haram Activity in the area around a school, and the Socioeconomic characteristics of the region around a school. Third, we train several predictive machine learning models and assess their predictive efficacy. The results show that some of these models can accurately quantify the likelihood that a school will be at risk of a Boko Haram attack. In addition, they cast light on the features that are most important in making such predictions. We then analyze learned decision trees to identify some conditions on the independent variables that help predict Boko Haram attacks on school. Fourth, we use these decision trees to formulate multivariate hypotheses that we investigate further from a statistical perspective. We find that Security presence near schools, Activity of Boko Haram in regions, and the Socioeconomic factors characterizing the region a school is in are all significant predictors of attacks. We conclude with a policy recommendation.