Enhancing University Admissions through Scalable and Fair Machine Learning Models: A Case Study from Yemeni Universities

Abstract

Machine learning (ML) techniques have been used to improve university admissions in Yemen, which faces major challenges. There is a bias against disadvantaged applicants. This bias comes from relying too much on traditional academic metrics. At the same time, weak technology leads to inefficiencies in handling many applications. These problems hurt the fairness and ability of current admission systems. To address this, we apply ML techniques. Various models, including logistic regression, support vector machines, k-nearest neighbors, random forest, and gradient boosting were evaluated. Their performance was evaluated based on accuracy and fairness. Among these, gradient boosting achieved the highest accuracy of 92% along with a notable 20% bias reduction measured by the Disparate Impact Ratio (DIR) and the Equal Opportunity Difference (EOD). Although all ensemble models demonstrated superior scalability and fairness compared to traditional methods. These models effectively processed datasets of over 15,000 student records while maintaining performance. Unlike complex deep learning models, the proposed models are easier to interpret. They show the clear importance of the feature. Entrance exam scores and high school GPA are the main predictors. Furthermore, the framework incorporates fairness-aware optimization, effectively reducing biases in admission decisions and enhancing suitability for socioeconomically diverse applicant pools. The findings illustrate that ML-driven approaches can revolutionize university admissions by delivering scalable, equitable, and resource-efficient solutions, which are particularly beneficial for institutions in constrained environments.