AUTOMATED STUDENT ENGAGEMENT DETECTION USING HYBRID DEEP LEARNING MODELS

Student classroom behavior represents a critical indicator of learning effectiveness and pedagogical success, yet traditional assessment methods remain subjective and resource-intensive.This paper introduces an innovative computer vision-based framework for automated student behavior detection utilizing a hybrid deep learning architecture that combines YOLOv8 object detection with LSTM temporal classification.We established a comprehensive behavioral annotation protocol with rigorous inter-annotator reliability assessment, achieving a Fleiss' Kappa score of 0.76 across three trained annotators.The proposed hybrid model integrates specific behavioral detection with educational psychology theory to classify distinct student behaviors in real-time classroom environments.Our methodology addresses fundamental challenges in educational technology by providing objective, scalable behavior assessment tools.The systematic experimental framework ensures reproducibility and validity through ethical approval, standardized data preparation, and multi-scenario evaluation protocols.