Page 124 - 2024-Vol20-Issue2

120 | Assim & Mahmood

TABLE II.
EPILEPTIC DETECTION FROM EEG SIGNALS (Continued)

Ref Objective Dataset Preprocessing Classification Model Tools Results

Using the Random forest

To achieve high Pauta criterion, model

accuracy, reduce the impact The outcome Accuracy
=91.78%,
sensitivity, of noise. ICA4 for is determined sensitivity
=91.27%,
and filtering out 95% by voting
and
[44] specificity CHB–MIT. of the noise. or averaging N/A. specificity
=93.61%.
in EEG Analysis of the results

classification variance by optimized

using phase P-value method. parameters by

synchronize. Phase using the

Synchronization. grid search.

To achieve CNN with Accuracy
in a single
[45] high accuracy CHB–MIT. The short LeNet-5, N/A. channel is
in EEG temporal Fourier the network 86%. In
transforms (STFT) Two pooling Matlab multichannel
classification layers and N/A. the accuracy
using short contain time two convolutional increased
temporal Fourier and frequency. layers make up N/A to 90%.
transforms PC with NVIDIA Accuracy for

(STFT). the model. Titan XP Pro binary
GTX1080Ti classification
To achieve AlexNet CNN 12 GB GPU,
1 TB HDD, and = 100%
high accuracy in A spectrogram model 8 GB RAM with and for
an Intel Core i7 ternary
binary and is used to Convolutional 3.90 GHz CPU. classification
= 100%
[46] ternary EEG Bonn translate neural network classification
classification University. the EEG signal in two accuracy
in the
using into visual dimensions 2class
=99.95%,
spectrogram data. and the idea of 3class
=99.98%,
data. transfer learning. 4class
=99.96%,
To achieve Splitting EEG Epileptic-Net
signal with model which and
high classification a set size 5class
window into integrates = 99.96%.
[47] accuracy in Bonn DCB, FAM,
multiple classes Universitys. several smaller RB, and HT
signals.
using the Adam
optimizer.
Epileptic-Net

model.

To achieve 3D-CNN.
Three distinct
high accuracy Oversampling
method, CNNs are
in EEG built to
Sliding window,
[48] classification CHB-MIT. FFT, and separate deep N/A Accuracy =
using WPD. and beginning 98.33±0.18

oversampling, features.

sliding window,

FFT.