Utilizing Fair and Explainable Machine Learning to Analyze High School Graduation Likelihood

Abstract

The traditional focus of data-driven decision-making has been business applications. Other domains, like education, also have significant potential. For instance, various factors impact the likelihood of successfully graduating; advance identification of students "at risk" of not graduating allows administrators to intervene, increasing graduation likelihood. This Research-Practice Partnership applies Machine Learning (ML), including aspects of Fairness and Explainability, to identify High School students at risk of not graduating. We show that ML approaches can successfully predict such students, while Explainable ML techniques can shed light on the factors that contribute most to a reduced likelihood of graduation. With this information, school counselors can efficiently identify roadblocks and also follow up with "grey zone" students, i.e., students at risk of not graduating but who do not follow typical non-graduation patterns (and might not be on the radar of counselors).