Federated learning-enabled privacy-preserving framework for seizure forecasting and affective state analysis using multi-modal EEG-ECG data

Seizure forecasting and affective state analysis using EEG-ECG data play a pivotal role in advancing neurological and mental health monitoring. However, existing methods such as Fed-Transformer, Res-1D CNN, and Fed-ESD suffer from privacy risks, inefficient feature extraction, and high computational overhead, limiting their effectiveness in real-world applications. To overcome these challenges, this study proposes NeuroFedSense, a novel Federated Learning-enabled Privacy-Preserving Framework that integrates a Temporal Convolutional Network (TCN) with an Attention Mechanism for accurate seizure forecasting and affective state analysis using EEG-ECG data, ensuring enhanced feature selection, interpretability and efficient decentralized training. The model leverages adaptive attention-based optimization and weighted feature selection to improve classification performance while ensuring data privacy. Implemented using TensorFlow, NeuroFedSense achieves 99.54% accuracy, 99.62% precision, 99.34% recall, and a 99.46% F1-score, outperforming Fed-Transformer (97.10% accuracy), Res-1D CNN (81.62% accuracy), and FML (99.10% accuracy). The ROC-AUC score of 0.99 further establishes its superiority over competing models. Additionally, the federated approach reduces energy consumption per node by 30% and optimizes communication efficiency by minimizing data transmission by 15% over 100 rounds. By ensuring high accuracy, improved privacy, reduced computational overhead, and enhanced energy efficiency, NeuroFedSense sets a new benchmark for decentralized, real-time seizure prediction and affective state monitoring. These findings underscore its potential for deployment in intelligent, privacy-preserving healthcare applications, addressing critical challenges in remote neurological monitoring.