This research uses machine learning-based AI models to predict admissions decisions at a large urban research university. Admissions data spanning five years was used to create an AI model to determine whether a given student would be directly admitted into the School of Science under various scenarios. During this time, submission of standardized test scores as part of a student's application became optional which led to interesting questions about the impact of standardized test scores on admission decisions. We first developed AI models and analyzed these models to understand which variables are important in admissions decisions, and how the decision to exclude test scores affects the demographics of the students who are admitted. We then evaluated the predictive models to detect and analyze biases these models may carry with respect to three variables chosen to represent sensitive populations: gender, race, and whether a student was the first in his family to attend college.

Bias and fairness issues in artificial intelligence (AI) algorithms are major concerns, as people do not want to use software they cannot trust. Because these issues are intrinsically subjective and context-dependent, creating trustworthy software requires human input and feedback. (1) Introduction: This work introduces an interactive method for mitigating the bias introduced by machine learning models by allowing the user to adjust bias and fairness metrics iteratively to make the model more fair in the context of undergraduate student admissions. (2) Related Work: The social implications of bias in AI systems used in education are nuanced and can affect university reputation and student retention rates motivating a need for the development of fair AI systems. (3) Methods and Dataset: Admissions data over six years from a large urban research university was used to create AI models to predict admissions decisions. These AI models were analyzed to detect biases they may carry with respect to three variables chosen to represent sensitive populations: gender, race, and first-generation college students. We then describe a method for bias mitigation that uses a combination of machine learning and user interaction. (4) Results and Discussion: We use three scenarios to demonstrate that this interactive bias mitigation approach can successfully decrease the biases towards sensitive populations. (5) Conclusion: Our approach allows the user to examine a model and then iteratively and incrementally adjust bias and fairness metrics to change the training dataset and generate a modified AI model that is more fair, according to the user’s own determination of fairness.

Bias detection and mitigation is an active area of research in machine learning. This work extends previous research done by the authors to provide a rigorous and more complete analysis of the bias found in AI predictive models. Admissions data spanning six years was used to create an AI model to determine whether a given student would be directly admitted into the School of Science under various scenarios at a large urban research university. During this time, submission of standardized test scores as part of an application became optional which led to interesting questions about the impact of standardized test scores on admission decisions. We developed and analyzed AI models to understand which variables are important in admissions decisions, and how the decision to exclude test scores affects the demographics of the students who are admitted. We then evaluated the predictive models to detect and analyze biases these models may carry with respect to three variables chosen to represent sensitive populations: gender, race, and whether a student was the first in his or her family to attend college. We also extended our analysis to show that the biases detected were persistent. Finally, we included several fairness metrics in our analysis and discussed the uses and limitations of these metrics.