A Systematic Review of Educational Facial Emotion Recognition: Datasets, Methods, Modality, and Potential Transfer to Vocational Teaching Contexts

Abstract

Facial emotion recognition (FER) has emerged as a promising component of educational technology, yet its integration into pedagogical practice remains uneven, particularly in vocational learning contexts. This systematic review examines 38 empirical studies published between 2015 and 2025, identified through a PRISMA-guided search of major academic databases, including Scopus. The synthesis explores how datasets, model architectures, multimodal learning signals, and system design shape the applicability of affect analytics in authentic instructional settings. The findings indicate that progress in educational FER is largely driven by benchmark datasets and incremental architectural refinement evaluated under controlled conditions, which limits transferability to hands-on learning environments. While lightweight, attention-enhanced models and multimodal approaches improve deployment feasibility and affective interpretation, most systems remain open-loop and rarely support sustained pedagogical adaptation. Overall, the review highlights that advancing educational FER requires closer alignment between data practices, model and modality design, and the pedagogical realities of vocational education.