Insights into University Composite Rankings from Explainable AI Counterfactuals

Abstract

University rankings exert considerable influence in higher-education decision-making. Yet, as an artifact of their construction, rankings are largely unhelpful in conveying practical strategic insights to university administrators’ intent on improving their college’s rank. Machine learning tools such as interpretable machine learning (IML) and explainable artificial intelligence (XAI), taking aim at piercing obscure, black-box algorithms have gained a lot of interest recently. However, there appear to be few deployments of their use in appraising University rankings. In this work, using as representative dataset the QS Rankings of USA MBA programs, we show how counterfactual XAI can support proactive responses by educational stakeholders to rankings outcomes.  Explaining individual predictions opens great opportunities for intervention and strategizing. The method is applicable to any extant rankings.