Machine Learning-Enabled Quality Analytics for Higher Education: A Multi-Dimensional Classification and Decision-Support Framework

Abstract

Higher education institutions generate heterogeneous information through curriculum processes, teaching-learning activities, learning resources, student support, research and extension, internal quality assurance, and learner engagement. Converting these heterogeneous signals into actionable quality intelligence is a computationally challenging classification problem because educational quality is multidimensional, noisy, partially observed, and context-dependent. This paper proposes and experimentally evaluates a machine-learning quality analytics pipeline to classify higher-education quality states. We constructed a computational dataset of 500 synthetic student-level observations using seven normalized quality dimensions. Three supervised learning algorithms Logistic Regression, Decision Tree and Random Forest were evaluated using a stratified 70:30 hold-out protocol and five-fold cross-validation. Performance was assessed through accuracy, precision, recall, F1-score and ROC-AUC. Random Forest achieved 69.33% hold-out accuracy and an F1-score of 79.46%, while Logistic Regression achieved 69.33% accuracy and 79.28% F1. We also used five-fold cross-validation to assess performance stability. Feature-importance analysis identified the relative contribution of the quality dimensions, and an ablation experiment examined the effect of removing the most influential predictor. The study also develops a probabilistic feedback-validation formulation and an outcome-oriented Faculty Effectiveness Index architecture for integration into a quality-management pipeline. The results demonstrate the feasibility of treating institutional quality management as a computational decision-support problem rather than as a collection of isolated descriptive indicators. Because the numerical dataset is synthetic, the reported performance values are methodological demonstrations rather than estimates of any real institutional population. The framework provides a technically explicit basis for subsequent validation with ethically collected institutional data.