Data mining approach in seizure detection

Extracting the meaningful patterns from time series dataset is one of the interesting and challenging task of data mining especially from time series brain dataset. A seizure can be detected either by EEG or ECoG signal. The applied ECoG dataset is of sample size 400Hz and it has two class values one is ictal (seizure) another is pre-ictal (before seizure). The previous study is not much sufficient to classify the state of seizures on the ground of classifier decision forest. In this paper, we classify the seizure states and motivates the importance of electrodes which are placed on the brain surface. Our analysis suggests through the systematic forest (SysFor) a type of decision forest that some electrodes are more fruitful to detect the seizure and found two features `min' and `max' are the best features. The pattern also reveals three trees detect seizure if the root attribute values are less than the value of selected attribute, and seven trees detect the seizure if root attribute values are greater than the selected value of root attribute.