Implementing federated learning for privacy-preserving emotion detection in educational environments.

Emotion detection has become an essential tool in educational settings, where understanding and responding to students' emotions is crucial to improving their engagement, academic performance, and emotional well-being. However, traditional emotion detection systems, such as DeepFace, and hybrid transformer-based models face significant data privacy and scalability limitations. These models rely on transferring sensitive data to central servers, compromising student confidentiality and making deployment in large or diverse populations difficult. In this work, we propose a federated learning-based model designed to detect emotions in educational settings, preserving data privacy by processing them locally on students' devices (smartphones, tablets, and laptops). The model was integrated into the Moodle platform, allowing its evaluation in a conventional educational environment. Advanced anonymization and preprocessing techniques were implemented to ensure the security of emotional data and optimize its quality. The results demonstrate that the proposed model achieves a precision of 87%, a recall of 85%, and an F1-score of 86%, maintaining its performance under adverse conditions, such as low lighting and ambient noise. In addition, a 15% increase in academic participation and a 12% improvement in the average academic performance of students were observed, highlighting the system's positive impact on educational dynamics. This innovative method combines privacy, scalability, and performance, positioning itself as a viable and sustainable solution for emotion detection in contemporary educational environments.