Artificial Intelligence–Enabled Digital Transformation in Higher Education: A Data-Driven Framework for Addressing Learning Inequalities, Administrative Inefficiencies, and Barriers to Institutional Innovation

Abstract

Higher education institutions increasingly adopt digital technologies, yet learning inequalities, administrative delays, and organizational barriers continue to restrict their transformative potential. This study develops an artificial intelligence enabled framework for integrating equitable student support, administrative optimization, and institutional innovation at the National University of Sciences and Technology (NUST), Pakistan. The dataset comprises 120,000 student learning records, 18,000 administrative service requests, and survey responses from 2,400 students and 360 staff members collected over three academic years. Recorded variables include academic performance, socioeconomic circumstances, digital accessibility, learning engagement, service processing times, technology adoption, and institutional readiness. The proposed framework integrates XGBoost to identify students requiring additional academic support, natural language processing to classify administrative requests, and interpretable clustering to characterize institutional innovation barriers. Data from earlier academic periods are used for training and validation, while the latest period supports independent testing. Student identifiers are separated across partitions to minimize information leakage. Evaluation considers predictive discrimination, classification accuracy, subgroup sensitivity, administrative processing time, and institutional readiness, using logistic regression and conventional service routing as comparative baselines. Experimental results show that XGBoost achieves an area under the receiver operating characteristic curve of 0.91 and a macro F1 score of 0.86, exceeding logistic regression values of 0.82 and 0.76, respectively. Following targeted academic support, the completion gap between digitally disadvantaged and advantaged learners decreases from 14.2 to 9.1 percentage points. Automated administrative request classification achieves 92.4% accuracy, while median processing time declines from 6.8 to 4.3 days. Clustering identifies inadequate infrastructure, limited staff capabilities, and fragmented governance coordination as principal barriers to institutional innovation. Following coordinated interventions, the institutional readiness index increases from 58.6 to 72.3 on a normalized scale. The findings indicate that integrated artificial intelligence can support institutional decisions concerning educational inclusion, administrative responsiveness, and innovation planning. However, observed changes require cautious interpretation because predictive performance and intervention outcomes do not independently establish causality. The framework incorporates data minimization, explainable predictions, human oversight, and subgroup auditing. Its contribution is a practical analytical architecture linking student needs, operational performance, and organizational readiness within higher education.