EEG Time-Series clustering using ARMA Modeling and the Grassmannian
MetadataShow full item record
The detection of epileptic seizures is of primary interest for the diagnosis of patientswith epilepsy. Epileptic seizure is a phenomenon of rhythmical discharge for eithera focal area or the entire brain and this individual behavior usually lasts fromseconds to minutes. The unpredictable and rare occurrences of epileptic seizuresmake the automated detection of seizures highly recommended especially in longterm EEG recordings.Technological advancement in noninvasive methods such as Electroencephalographic(EEG) helped doctors to analyze hour long EEG recordings to diagnoseif patient is suffering from epilepsy or not. To help physicians diagnose seizure,we developed automatic seizure detection algorithm. We use model based featureextraction method, Auto Regressive Moving Average (ARMA) and use recently developedhypothesis from Riemannian Multi-Manifold Modeling to extract featuresin the Grassmannian. Later we use one of the hierarchical-clustering algorithm,Louvain method, to cluster the time series.This method is evaluated using real University of Bonn EEG dataset. Simulationresults were promising as compare to State of Art clustering methods andoutperformed them. An advantage of this method that no training data are useddue to the unsupervised nature of clustering.