Classification of Normal and Epileptic Seizure EEG Signals Based on Empirical Mode Decomposition - Complex System Modelling and Control Through Intelligent Soft Computations - page 378

Table 2 Comparison of the proposed method for classi cation of normal and epileptic seizure
EEG signals with the existing methods studied on same dataset
Authors Method Accuracy

(%)

Nigam and Graupe
( 2004 )

Nonlinear pre-processing filter and diagnostic neural
network

97.2

Srinivasan et al.
( 2005 )

Time and frequency domain based features and
recurrent neural network

99.6

Kannathal et al.
( 2005 )

Entropy based measures and adaptive neuro-fuzzy
inference system

about 90

Polat and G ü ne ş
( 2007 )

Fast Fourier transform based features and decision tree 98.72

Subasi ( 2007 ) Discrete wavelet transform based features and mixture

of expert model

94.5

Tzallas et al. ( 2007 ) Time-frequency analysis based features and arti cial

neural network

100

This work Proposed method 100

measures used in this work are the simple and can be used as indicators for
diagnosis of epilepsy. Moreover, these parameters are de ned in time domain

which can help us to implement the proposed methodology for epileptic seizure
detection with low computational complexity. It can be observed that performance
of the proposed method in terms of accuracy is better than that of the other com-
pared methods. The experimental analysis of the proposed method shows that
features based on area measures are very effective to represent the behavior of
epileptic seizure EEG signals giving excellent classi cation performance.

4 Conclusion

This topic chapter has developed a novel approach for classi cation of the normal

and epileptic seizure EEG signals using empirical mode decomposition and com-
puting two area parameters for IMFs. Since the EEG signal is non-linear and non-
stationary in nature, the EMD which is data dependent approach and suitable for
analysis of nonlinear and non-stationary signals, ef caciously decompose the EEG

signals into IMFs which are oscillatory components. In this work, we have explored
the capability of two area parameters as the features for classi

cation of normal and
epileptic seizure EEG signals. It is noteworthy that the symmetric nature of IMFs,
makes it possible to compute these two area measures and justi es the application of

EMD before feature extraction from EEG signals. Computation of area measures
uses the analytic signal representation of IMFs and SODP of IMFs. The IMFs have
single center of rotation with circular geometry in analytic signal representation.
Similarly, IMFs also exhibit elliptical patterns in SODP. Consequently,

these