Emotion-Aware Multimodal Learning Analytics for Early Prediction of At-Risk Learners Using Hybrid Machine Learning

Identifying at-risk learners is a critical challenge in education, affecting students' cognitive and emotional well-being and influencing learning outcomes. The predictive models depend on academic indicators, including affective factors such as engagement, confusion, and frustration. This study proposed an emotion-aware multimodal learning analytic framework. It incorporates the academic and emotional features within a hybrid machine learning architecture. The pilot study used academic performance data, combined with emotional indicators, to evaluate the feasibility of the approaches. The framework works with K-Means clustering for learner profiling and Random Forest Classification with cross-validation for predictive modeling. The automated feedback mechanism predicts with adaptive instructional strategies. Thus, it demonstrates the emotional and academic indicators that improve the robustness and provide insights into learner behavior. Thus, the proposed framework provides a scalable, interpretable framework across educational contexts. Keywords Learning Analytics; Emotion-aware ; Adaptive instructional; Multimodal Learning; Hybrid Machine Learning; Explainable AI