Development of tuned hybrid fuzzy and BiLSTM-based epileptic seizure classification model with stacked 1D-CNN-fisher discriminant feature selection

ABSTRACT

Electroencephalogram (EEG) is a signal which consists of different sinusoidal components with a dense frequency spectrum. The existing methods are time-consumingand includes inconsistencies in judgment among seizure classification. Therefore, it is very difficult to accurately detect epileptic seizures from the recorded EEG. Therefore, a new epileptic seizure detection model is developed with hybrid deep-structured technology. The EEG signal data obtained from the standard online resources which are given into the pre-processing phase with the help of band pass filtering and smoothing techniques. Then, the 5-level Discrete Wavelet Transform (5-level DWT) is used for signal decomposition to get the decomposed signals. 1-D stacked Convolutional Neural Network (1-D stacked CNN) is utilized for the extraction of features. After, the feature selection process is executed by utilizing the Fisher Discriminant Analysis (FDA). The selected features are subjected to the classification phase by Tuned Hybrid Fuzzy Bi-directional Long Short-Term Memory (THFBi-LSTM). Here, the parameter optimization takes place with hybridized optimization algorithm of Probability-based Dingo Coyote Optimization (P-DCO). In the overall simulation estimation, the offered approach achieves a 97% accuracy rate and also a 92% F1-score rate. Thus, the experimental results are reveled that it is better than the other baseline approaches.

KEYWORDS:

• Epileptic Seizure Classification
• Tuned hybrid fuzzy and bi-directional long short term memory
• Probability-based dingo coyote optimisation
• 5-level discrete wavelet transform
• 1-D stacked convolutional neural network

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notation Table

$P_{u,v}$	=	EEG signal output
$F_b^{ext}$	=	The output of feature extraction
$X_{M\times N}$	=	Filter matrix
$F_c^{slt}$	=	The output of feature selection
$n_q^k$	=	Center of the membership function
${\beta }_{m,q}^Q$	=	The right spread
${\beta }_{m,q}^P$	=	Right spread of the membership function
${\vec{f}}_m$	=	Hidden output signal
$B{L}_i$	=	The lower bound of the search space
$P{b}_{sct}$	=	Probability of scattering
$P{b}_{asn}$	=	The probability of association
$\left[2,\frac{po{p}_{size}}{2}\right]$	=	The interval of random number
$po{p}_{size}$	=	The total population size of the dingoes
${\vec{h}}_m\left(k+1\right)$	=	The movement of the dingoes
$E_a^{co}$	=	Optimized epochs in the Bi-LSTM classifier
$Ex{B}_c^{DO}$	=	The optimized exponential bound in Fuzzy

Correction Statement

This article has been republished with minor changes. These changes do not impact the academic content of the article.