Privacy-aware federated machine learning algorithms for analyzing students' psycho-emotional state data

Artificial Intelligence (AI) and Machine Learning (ML) have transformed industries by enabling intelligent decision-making and predictive analytics using big data. This data, defined by its volume, velocity, and variety, drives complex ML models that deliver accurate insights. However, traditional centralized ML approaches, which require collecting data on central servers, raise significant privacy and security concerns. Federated Learning (FL) proposes a decision by providing decentralized model training. is especially valuable for analyzing sensitive data, such as student’s psycho-emotional states, while addressing challenges like data heterogeneity and limited device resources. Our research focuses on FL algorithms, including FedAVG and FedOPT, FedProx applied to psycho-emotional state analysis in educational settings. Our findings reveal that FedProx achieves the highest accuracy among the FL algorithms in analysis of psycho-emotional state. Our comparison between FedAVG and FedOPT, FedProx show that they are practically equivalent after 10 rounds. By addressing data variability, infrastructure limitations, and privacy concerns, this study enhances FL frameworks for analyzing academic stressors and their impact. These frameworks provide accurate insights into students’ well-being while ensuring data privacy, thereby supporting the development of nurturing learning environments.