Digital Twin-Based Learning Analytics with Fog Computing and LLAMA

Real‑time learning analytics in higher education are often constrained by the latency, bandwidth, and privacy limitations of cloud‑only architectures, which hinder the de-livery of timely, actionable feedback; this study addresses that gap. We introduce Learner’s Digital Twin, a framework that integrates fog computing at the network edge with Meta‑LLAMA to interpret multimodal student data and provide instant, personalized feedback and educator insights. The architecture performs local processing on fog nodes to reduce delay and limit data movement, while LLAMA generates context‑aware text analyses; predictive components include linear regression to forecast final‑exam scores from attendance, assignment averages, and participation, and K‑means clustering to profile learning patterns. We evaluated the framework in a real educational setting over three months, using Postman‑based latency tests and user surveys. The system reduced average response latency by ~300ms. The feedback generated was personalized, and survey responses indicated positive user perceptions: for students, 80% reported overall satisfaction, with >90% perceiving the feedback as personalized and >75% finding it relevant; teachers similarly reported ~80% satisfaction. These findings indicate that combining a digital‑twin paradigm with fog computing and LLM can support timely, personalized feedback and actionable insights in high-er‑education contexts; future work should examine scalability and generalizability across diverse settings.