An Adaptive Neuro-Fuzzy Interface System for Classifying Sleep EEG

Document Type : Original Article

Abstract

In the present paper, classification of sleep stages of EEG by using Adaptive Neuro -Fuzzy. Six sleep EEG records for
each of ten patients were selected from Cairo Canter of Sleep Disorder. Three methodologies of analysis were util ized
for feature extraction. These include: autoregressive modelling (AR), bispectral analysis, and discrete wavelet transform
(DWT), where principle component analysis (PCA) was used to reduce feature dimensionality. The features derived
from the three methodologies of feature extraction were used as input feature vectors to the classifier. The classification
rates reached are 89.5%, 92% and 90.8% for the AR modelling, the bispectral analysis, and DWT, respectively. To
improve the classification accuracy a data fusion at the matching score was utilized. The total classification accuracy
reached 94.3%.

Keywords