Research On Predicting Abnormal Students Behavior In A Learning Environment With Poor Data

With the advancement of technology innovation, the evolution of remote inspection platforms and monitoring via camera systems or webcams has seen significant and rapid growth. Such rapid development posed a significant challenge in ensuring fairness in evaluating learner performance. Conventional AI models require extensive training datasets, which becomes particularly challenging in online environments with unstable conditions and numerous participants (each participant will have different webcam parameters and external conditions). To address these constraints, we propose a method to predict abnormal facial behavior in learners with limited data. Our approach involves gathering data by extracting images from videos provided by subjects during sessions prior to assessments. These images are categorized by objects and behaviors, then annotated accordingly. Through rigorous testing and real-time evaluation, our model achieves a robust 98% accuracy in detecting learner behaviors. This innovative solution leverages limited datasets effectively, showcasing its potential to enhance assessment fairness and efficiency in dynamic educational settings.