GENERATIVE AI-POWERED ADAPTIVE LEARNING FOR PERSONALIZED HIGHER EDUCATION: MULTIMODAL LEARNING ANALYTICS AND EARLY PREDICTION OF ACADEMIC FAILURE IN PAKISTANI UNIVERSITIES

Abstract

The convergence of Generative Artificial Intelligence (GenAI) and Multimodal Learning Analytics (MMLA) represents a transformative paradigm shift in higher education, enabling personalized adaptive learning and early prediction of academic failure. This review paper provides a comprehensive synthesis of GenAI-powered adaptive learning frameworks integrated with MMLA and Early Warning Systems (EWS), specifically contextualized within the socio-technical realities of Pakistani higher education institutions. The paper examines the paradigmatic evolution from rule-based Intelligent Tutoring Systems to dynamic, context-aware generative architectures that synthesize hyper-personalized content and simulate interactive tutoring dialogues in real-time. It systematically analyzes MMLA data fusion methodologies early, late, and semantic fusion that aggregate heterogeneous data streams including LMS clickstreams, textual discourse, and behavioral engagement metrics to construct comprehensive student learning profiles. The review evaluates machine learning algorithms and Explainable AI frameworks, particularly SHAP-based interpretability, for predicting academic failure risk with demonstrated classification accuracy reaching 84.5% and ROC-AUC values exceeding 0.902. Critically, the paper examines the structural realities of Pakistani universities under Higher Education Commission (HEC) governance, identifying infrastructural deficiencies, digital divides, and faculty literacy gaps as significant implementation barriers. The analysis engages with Pakistan's emerging regulatory frameworks, including the National AI Policy 2025 and HEC Draft Framework on Ethical Generative AI, highlighting the academic integrity paradox posed by AI detection tools' documented biases against non-native English writers. The paper proposes a six-stage Pakistan AI Education Translation Framework addressing infrastructure stabilization, data privacy, teacher readiness, curriculum redesign, assessment reform, and evidence-based scaling. This review contributes both theoretical insights and practical implementation guidance for integrating GenAI and MMLA in resource-constrained higher education contexts.