Epileptic EEG signal classification using an improved VMD-based convolutional stacked autoencoder

Abstract

Numerous techniques have been explored so far for epileptic electroencephalograph (EEG) signal detection and classification. Deep learning-based approaches are in recent demand for data classification with huge features. In this paper, an improved deep learning approach based on convolutional features followed by stacked autoencoder (CSAE) and kernel extreme learning machine (KELM) classifier at the end is proposed for EEG signal classification. The convolutional network extracts initial features by convolution, and after this stage, the features are supplied to stacked autoencoder (SAE) for obtaining final compressed features. These suitable features are then fed to KELM classifier for identifying seizure, seizure-free and healthy EEG signals. The EEG signals are decomposed through chaotic water cycle algorithm-optimised variational mode decomposition (CWCA-OVMD) from which the optimised number of efficient modes is obtained yielding six features like energy, entropy, standard deviation, variance, kurtosis, and skewness. These CWCA-OVMD-based features are then fed to the CSAE for the extraction of relevant features. Once the features are obtained, the KELM classifier is used to classify the EEG signal. The classification results are compared with different deep learning classifiers validating the efficacy of the proposed model. The KELM classifier avoids the choice of hidden neurons in the end layer unlike traditional classifiers which is one of the major advantages.

Appendices

Appendix:

A. Data normalisation between [0, 1].

$$ d_{{\text{n}}} = \frac{{d - d_{{{\text{min}}}} }}{{d_{\max } - d_{{{\text{min}}}} }} $$

(A.1)

where $d_{{\text{n}}}$ = normalised EEG data, d = EEG data.

$d_{{{\text{min}}}}$ = minimum value of EEG data.

$d_{\max }$ = maximum value of EEG data.

B. Performance evaluation

Classification accuracy (AC), sensitivity (SN), and specificity (SP) are defined below:

$$ {\text{AC}} = \frac{{P_{{\text{T}}} + N_{{\text{T}}} }}{{P_{{\text{T}}} + N_{{\text{T}}} + P_{{\text{F}}} + N_{{\text{F}}} }} \times 100\% $$

(B.1)

$$ {\text{SN}} = \frac{{P_{{\text{T}}} }}{{P_{{\text{T}}} + N_{{\text{F}}} }} \times 100\% $$

(B.2)

$$ {\text{SP}} = \frac{{P_{{\text{T}}} }}{{P_{{\text{T}}} + P_{{\text{F}}} }} \times 100\% $$

(B.3)

where $P_{{\text{T}}}$ = True Positive,$N_{{\text{T}}}$ = True Negative, $P_{{\text{F}}}$ = False Positive, and $N_{{\text{F}}}$ = False Negative.

C. Feature extraction

The details of statistical features are described below:

$$ {\text{Energy:}}\;E = \sum\limits_{I = 1}^{N} ( d_{i} )^{2} $$

(C.1)

$$ {\text{Entropy:}}\,{\text{En}} = - \sum\limits_{I = 1}^{N} {p(} d_{i} )\log_{b} p(d_{i} ) $$

(C.2)

$$ {\text{Standard Deviation}}:{\text{St}} = \sqrt {\frac{{\sum\limits_{I = 1}^{N} {} (d_{i} - \mu )^{2} }}{N}} ,\;\mu = \frac{1}{N}\sum\limits_{I = 1}^{N} {d_{i} } $$

(C.3)

$$ {\text{Variance}}:{\text{Va}} = \frac{1}{N}\sum\limits_{I = 1}^{N} {(d_{i} } - \mu )^{2} $$

(C.4)

$$ {\text{Kurtosis}}:{\text{Ku}} = \frac{1}{N}\sum\limits_{I = 1}^{N} {\left( {\frac{{d_{i} - \mu }}{{{\text{St}}}}} \right)^{4} } $$

(C.5)

$$ {\text{Skewness}}:\;{\text{Sk}} = \frac{1}{N}\sum\limits_{I = 1}^{N} {\left( {\frac{{d_{i} - \mu }}{{{\text{St}}}}} \right)^{3} } $$

(C.5)

$d_{i}$ = ith input data, $N$ = Total number of samples, $p(d_{i} )$ = probability of input data, and $\mu$ = mean value of input data.