Statistical features based epileptic seizure EEG detection - an efficacy evaluation

Electroencephalographic (EEG) patterns are electrical signals generated in association with neural activities. Most anomalies in brain functioning manifest with their signature characteristics in EEG pattern. Epileptic seizure, which is a brain abnormality well-studied through EEG analysis, is an abnormal harmonious neural activity in the brain characterized by the presence of spikes in EEG. An automated detection of epileptic seizures proves useful to Neurologists in the diagnosis of epileptic patients. This work contributes towards the study of efficacy evaluation of statistical features towards classification of EEG data as Ictal, Inter-Ictal and Normal. The statistical features considered are energy, entropy, median absolute deviation, interquartile range, skewness and kurtosis. The features extracted from a real dataset of 500 time series, comprising of 100 Ictal, 200 Inter-Ictal and 200 Normal are given to classifiers such as Support Vector Machine(SVM), Fuzzy k-Nearest Neighbor (Fuzzy k-NN), k-Nearest Neighbor(k-NN) and Naive Bayes for three class classification. Each of the features were used separately for classification to determine their individual efficacies. Alongside, the popular feature ranking method `ReliefF´ has been used to rank the features. Both the evaluations resulted in “entropy” being ranked as the feature with maximum efficacy.