Low complexity algorithm for seizure prediction using Adaboost

This paper presents a novel low-complexity patient-specific algorithm for seizure prediction. Adaboost algorithm is used in two stages of the algorithm: feature selection and classification. The algorithm extracts spectral power features in 9 different sub-bands from the electroencephalogram (EEG) recordings. We have proposed a new feature ranking method to rank the features. The key (top ranked) features are used to make a prediction on the seizure event. Further, to reduce the complexity of classification stage, a non-linear classifier is built based on the Adaboost algorithm using decision stumps (linear classifier) as the base classifier. The proposed algorithm achieves a sensitivity of 94.375% for a total of 71 seizure events with a low false alarm rate of 0.13 per hour and 6.5% of time spent in false alarms using an average of 5 features for the Freiburg database. The low computational complexity of the proposed algorithm makes it suitable for an implantable device.