Publication

Abstract : Automated detection of neonatal seizures presents a crucial clinical need, as prompt diagnosis is essential for optimizing neurodevelopmental outcomes. The comprehensive workflow of our neonatal seizure detection system begins with raw EEG acquisition, progresses through signal preprocessing and feature extraction stages, and culminates in model development and evaluation using a publicly available neonatal EEG dataset comprising 79 multi-channel recordings. This study presents a machine learning system utilizing Gradient Boosting that demonstrates exceptional performance in seizure identification, achieving 97.8% accuracy and a 96.5% F1-score on clinical EEG data. A comparative analysis of three machine learning approaches Random Forest, Multilayer Perceptron, and Gradient Boosting revealed that Gradient Boosting outperformed the others while maintaining high specificity 93.5% and significantly reducing false alarms. The system’s ability to reliably detect subtle seizure patterns, combined with its computational efficiency, makes it particularly suitable for deployment in neonatal intensive care units. By addressing a critical gap in neonatal neurology, this technology offers a practical solution for resource-limited settings where specialist expertise may be unavailable, potentially enabling earlier interventions and improving patient outcomes through timely seizure management.