<p>Real-time data analytics is essential for effective educational decision-making, yet traditional assessment systems often fail to capture students’ engagement and activity-based learning. This study proposes a real-time processing and visualization framework for educational big data, supported by edge intelligence. Using Internet of Things sensors across classrooms, labs, and learning environments, high-frequency multi-source student data, including attendance, lab activity, and interaction patterns, is captured and preprocessed with min–max normalization at the edge. The framework integrates a modified golden sine-driven adaptive layered LSTM (MGS-AL-LSTM) for behavior classification and a hidden Markov model for probabilistic decision-making, enabling low-latency, edge-based inference. A centralized dashboard provides real-time analytics and visualization, while selectively uploading normalized data to the cloud for long-term trend analysis. Performance improvements were validated using paired <i>t</i> tests and 95% confidence intervals across multiple experimental runs, confirming the robustness of gains. Simulations demonstrate that the proposed MGS-AL-LSTM model achieves 94% behavior classification accuracy, reduces decision latency by 48%, improves visualization clarity by 42%, lowers cloud upload by 41%, and enhances inference speed by 45% compared to baseline methods. When compared to the existing model, the proposed MGS-AL-LSTM model performs efficiently for processing large amounts of real-time educational data. For 200 users, the average response time is 0.8&#xa0;s, and the processing success rate reaches 99%, demonstrating the model's ability to maintain low latency and high reliability under high user load. These results illustrate that edge intelligence combined with advanced deep learning enables efficient, real-time educational analytics, empowering educators with timely insights to optimize teaching strategies and improve academic outcomes.</p> Graphical abstract <p></p>