Affective Computing in Online Higher Education: A Systematic Literature Review

Although affective states play a crucial role in education, they are often difficult to communicate and observe in online learning environments. This challenge has led to growing research on systems that can automatically detect affective states. This systematic literature review used PRISMA to analyze 96 studies on affective computing in online higher education, published between 2019 and 2024. The findings show that the most frequently studied affective states include learning-centered states, such as engagement, confusion, frustration, sentiment, as well as basic emotions, such as happiness, anger, sadness, surprise, and fear. Terminology often overlaps, and basic emotions are commonly used as proxies for learning-centered states. The most used modality is facial expression, with the dominant detection approach being deep learning, particularly convolutional neural networks. Most studies rely on self-collected datasets that, due to privacy concerns, are not publicly shared, limiting reproducibility and generalizability. FER2013, collected in a generic context, and DAiSEE, collected in an online educational setting, are the most used public datasets. A key limitation is that most systems are not evaluated in real classrooms, revealing a gap between technological development, and educational application. Ethical considerations are often overlooked, with privacy, when addressed, being the main focus. Finally, the review’s findings highlight the need for stronger integration between education and technology through interdisciplinary collaboration and real-world validation.