Page 123 - 2024-Vol20-Issue2

119 | Assim & Mahmood

TABLE II.
EPILEPTIC DETECTION FROM EEG SIGNALS

Ref Objective Dataset Pre-processing Classification Model Tools Results

CNN for Python

binary classification computer

To assess REPO Downscaled Raw 2-fully linked with two Accuracy for
the accuracy 2MSE EEG to 256 Hz and layers with a A 500 GB 3×131 model
of the model cohort split into 5-second single output of Memory,
[40] on ictal and overlapping pieces. = 0.869
are placed 32-core while
interictal after 3 blocks AMD EPYC 7551 the model
EEG of convolutional 3×5 accuracy
layers. Stochastic processors, = 0.825.
recordings. and Nvidia Tesla
Gradient T4 GPU. Keras

Descend SGD and TensorFlow

optimizer. v1.4. Are used.

Accuracy is:

CNN with originalEEG

To evaluate To divide 16 kernels Python/ =0.949
the signal and kernle NVIDI DWT
different into segments NVIDIA GeForce =0.9595,
using the size is RTX 2080 FFT=0.9371,
evaluation Hamming 31 × 1 STFT=0.943,
window The batch tests Hybrid
[41] techniques Bonn size is were run =0.9639
on the University with set to 6 EEG LSTM
length and 100 with =0.9787
accuracy 128. epochs. Keras 2.3.1. Hybrid
With Adam
of the

model.

optimizer. LSTM

=0.9908.

To achieve Standardize the 1D CNN Python Keras, DCAE+
raw input (3- blocks a package Bi-LSTM
high accuracy in EEG data of convolutional, built on model has
each block top of sensitivity
[42] 2-class, Bonn to a mean of consists of TensorFlow. =98.72%,
3-class, and University. 0 and a 5 layers, specificity
variation then 3 fully =98.86%,
5-class EEG of 1. connected accuracy
layers). Batch size =98.79%,
classification. =100 with F1-score
= 98.79%.
Adam
optimizer.

Signals are 2D-DCAE DCAE+
using 4 Bi-LSTM
filtered between models,
model
To achieve 0 and 128 Hz 2D-DCAE+MLP, has
2D-DCAE+
high and sampled Bi-LSTM, sensitivity
=98.72%,
accuracy, at 256 Hz. 2D-DCNN+MLP, Python specificity
2D-DCNN+ Google =98.86%,
[43] sensitivity, CHB–MIT. Use down Bi-LSTM) Colaboratory accuracy
specificity, sampling ., Adam =98.79%,
optimizer. F1-score
and F1-score to reduce = 98.79%.

in EEG dimensionality

classification. channel-by-

channel z-score

normalization.