A real-time multimodal attention monitoring system for online learning: integrating behavioral, affective, and activity indicators to support adaptive teaching

Introduction The fast growth of online and hybrid learning has increased the need for effective tools to monitor student attention during virtual lectures. Existing systems often depend on one or two models and do not combine behavioral, affective, and activity cues within a single structure. Methods This paper proposes a browser-based intelligent system to support adaptive teaching that refers to adjusting instruction according to students' attention levels .The proposed system integrates four complementary AI models for real-time attention monitoring without needing software installation. The system includes a behavioral features model based on Long Short-Term Memory (LSTM) for tracking face direction, head pose, hand movement, and mobile phone use, a drowsiness detection model based on the Eye Aspect Ratio (EAR) and Mouth Aspect Ratio (MAR), a facial expression model based on FaceAPI.js for detecting emotional states, lastly, an activity recognition model based on COCO-SSD and MediaPipe for detecting distracting behaviors. Results The individual models achieved accuracies ranging from 89.5% to 99.62%. Their outputs were then merged by using Majority Voting (MV), as a student was classified as inattentive when at least two models detected disengagement at the same time. Validation against a human expert achieved 96.67% accuracy, 100% precision, and a Cohen's Kappa of κ = 0.933, showing very strong agreement. Real-time alerts also supported timely instructor intervention, improving student attention from 30% to 95%, a gain of 65% points. Discussion The system provides two reports: a real-time dashboard that identifies inattentive students while the lecture and a post-lecture report that provides an overview of individual attention patterns. All processing is performed locally in the browser, with no student personal data stored or transmitted, thereby supporting privacy-preserving real-time attention monitoring.