A comparison of generative and discriminative approaches in automated neonatal seizure detection

Two systems based on different classifiers are compared for the task of neonatal seizure detection. Support vector machines and Gaussian mixture models are presented as examples of discriminative and generative approaches to classification. The performance of both systems is assessed using a number of metrics, the results of which indicate that both systems are competitive with other detectors in the literature. Finally, misclassified events are analysed, from which specific patterns affecting the performance of the detector are identified.