Fairness-Aware Predictive Analytics for Student Success in a Multicultural Higher Education Environment

Abstract

This research will seek to create a fairness conscious predictive analytics framework to learn and facilitate student achievement in a diverse university environment. In transnational education context predictive models have been shown to be an effective way of to identify at-risk students,  however, issues of algorithmic fairness and bias based on gender, nationality and scholarship status not yet been researched thoroughly. Anonymzed institutional data across multiple programs and semesters from a multicultural university context are used in this study. Predictive modeling is conducted using Logistic Regression and Gradient Boosting. Academic variables such as prior GPA and attendance are combined with demographic attributes, including age group, gender, nationality, and scholarship status. This study provides evidence of the importance of demographic variables relative to academic variables. This is critical for developing early warning systems, intervention strategies, and institutional policies in higher education organizations. The study is confined to four institutional setting and based on the accessible academic and demographic statistics. In order to maximize the strength, the data of four universities consisting of 16,500 students were involved in the analysis on the same set of features and modeling pipeline. University administrators and instructors to come up with data-driven and equitable intervention measures to address the needs of various learners can also use results. The fairness analysis framework has the potential to guide institutional policy about the responsible use of AI, early warning systems and specific academic support programs. This paper builds upon existing literature; it incorporates accuracy, interpretability, and equity in a single prediction analytics system designed to suit a transnational university environment.