An Advanced Deep Representation and Heuristic based Model for Reliable Epileptic Seizure Detection from Electroencephalogram Signals

An epileptic seizure is an unusual electrical flow in
brain that causes a temporary disturbance in normal brain
function and frequently results in convulsions or loss of
consciousness. An electroencephalogram (EEG) is the most critical
analytic device for epilepsy. Generally, the epileptic activity
recognition that depends on finding particular patterns in the
multi-modal EEG may be completed by a professional. The
detection of seizures on EEG data is a challenging task that
requires a relationship between clinical experts and machine
learning (ML) specialists to ensure the consistency and precision
of the detection approach for patients with epilepsy. Neural
network (NN)-based deep learning (DL) models like recurrent NN
(RNN) and convolutional NN (CNN) are applied on a labelled EEG
dataset that comprises non-seizure and seizure parts. This paper
proposes a complete and strong structure for epileptic seizure
recognition (ESR) employing EEG signals by relating
sophisticated DL techniques with a metaheuristic optimisation
model. Primarily, the EEG signals undergo an efficient pre
processing phase for normalising and transforming the raw data
into informative representations. Afterwards, a feature selection
(FS) approach is used to improve discriminative ability by
decreasing dimensionality. For seizure classification and
recognition, effective DL structures are used. At last,
metaheuristic-driven hyperparameter tuning is achieved to
further enhance the classification precision and robustness of the
system. Extensive experimental evaluation on benchmark EEG
datasets proves that the proposed structures significantly
outperform existing models.
Keywords—
Epileptic
Electroencephalogram
Hyperparameter Tuning
Signals;
Seizure;
Deep
Metaheuristics
I. INTRODUCTION
Learning;
Algorithm;