“Big Data” in Educational Administration: An Application for Predicting School Dropout Risk

Abstract

Purpose: In an era of unprecedented student measurement and emphasis on data-driven educational decision making, the full potential for using data to target resources to students has yet to be realized. This study explores the utility of machine-learning techniques with large-scale administrative data to identify student dropout risk. Research Methods: Using longitudinal student records data from the North Carolina Department of Public Instruction, this article assesses modern prediction techniques, with a focus on tree-based classification methods and support vector machines. These methods incorporate 74 predictors measures from Grades 3 through 8, including academic achievement, behavioral indicators, and socioeconomic and demographic characteristics. Findings: Two of the assessed classification algorithms predict high school graduation and dropping out correctly for more than 90% of an out-of-sample student cohort. Findings reveal a shift toward lower dropout incidence in regions hit hardest by the economic recession of 2008, especially for male students. Implications for Research and Practice: Machine-learning procedures, as demonstrated in this study, offer promise for allowing administrators to reliably identify students at risk of dropping out of school so as to provide targeted, intensive programs at the lowest possible cost.