Optimisation of an epileptiform activity detector for ambulatory use

A detector originally designed for seizure detection is modified to detect epileptiform activity in adults. The detector is intended for ambulatory use, and as such an emphasis is placed on the computational load of the detector. A framework is proposed making use of genetic algorithms in order to select the features for a Gaussian mixture classifier. Feature subset selection was performed by incorporating the computational load of each feature. This resulted in an improvement in classification results (larger area under both the ROC and PR curves), while reducing the runtime of the algorithm by up to 2000 fold with respect to a detector using the full feature set.