Facial expression recognition in learning environments: a review of deep learning advances, challenges, and future directions

Facial Expression Recognition (FER) has emerged as a pivotal research domain within affective computing, offering valuable insights into human emotions and affective behavior. In intelligent learning environments, FER supports adaptive feedback, personalized learning, and affect aware human AI interaction by leveraging learners' cognitive and emotional states to monitor their well-being. This review provides a comprehensive examination of recent deep learning advancements in FER, with a particular emphasis on its applications in education. It analyzes state-of-the-art architectures, including Convolutional Neural Networks (CNNs), Graph Neural Networks (GNNs), and transformer-based models, as well as emerging paradigms such as multi-label and multimodal learning. Benchmark datasets (e.g. RAF-DB, Aff-Wild2, MAFW) and evaluation protocols commonly used to assess model performance and generalization are also surveyed. Furthermore, the study highlights persistent challenges related to data scarcity, domain shift, fairness, and privacy, which continue to impede real-world deployment. To address these limitations, promising research directions and future opportunities are discussed. Overall, this review illustrates how the integration of image processing, psychology, and educational technology can drive the development of next-generation empathetic, context-aware learning systems capable of understanding and responding to complex human emotions.