FrAdadelta-CSA: Fractional Adadelta Chameleon Swarm Algorithm-based feature selection with SpikeGoogle-DenseNet for epileptic seizure detection

• Published in: Computational biology and chemistry Vol. 119; p. 108550
• Main Authors: Indu Salini, G., Sowmy, Sreeja, T.K.
• Format: Journal Article
• Language: English
• Published: England Elsevier Ltd 01.12.2025
• Subjects: Algorithms; Convolutional Neural Networks; Deep Learning; disease detection; Electroencephalography; Epilepsy - diagnosis; Epileptic seizure recognition; feature selection; Humans; Optimization; Seizures - diagnosis; Signal Processing, Computer-Assisted; NN; CNN; ESAASO; Epileptic seizure recognition; KHA; ELM; FM; DL; GBD; STFT; TTH; BN; Optimization; AUC; CSA; Deep learning; RNN; COA; GA; DM-ELM; feature selection; disease detection; LSTM; EEG; MOH; Adadelta-CSA; PSP; AM; PSO; Conv; DCSAE-ESDC; TL; LE; ReLU; FC; FrAdadelta-CSA
• ISSN: 1476-9271; 1476-928X; 1476-928X
• DOI: 10.1016/j.compbiolchem.2025.108550

In contemporary context, epileptic seizure stands as the most prevalent neurological disorder arising from the sudden atypical release of brain neurons and after effect of stress. The electroencephalogram (EEG) has been widely employed for epilepsy detection. Noteworthy are the Deep Learning methods and signal processing techniques utilized for seizures classification and detection. Through advancements in Deep Learning within the biomedical domain, several methodologies have been applied to identify and forecast seizure occurrences based on EEG data collected from individuals with epilepsy, typically confined from temporary in medical screening with standard scalp-EEG or intra-cerebral electrodes. This work aims to generate a mechanism for seizure recognition from EEG signals using a classification technique based on Deep Learning. The initial phase involves pre-processing, where denoising of input EEG signals is performed by employing the Short-Time Fourier Transform (STFT). Consequently, time-domain, spectral, and statistical features were extracted from pre-processed signals. Then, feature selection is performed utilizing Fractional Adadelta Chameleon Swarm Algorithm (FrAdadelta-CSA), a method that integrates the notion of fractional calculus into Adadelta Chameleon Swarm Algorithm (Adadelta-CSA). Finally, seizure prediction is conducted based on the selected features using SpikeGoogle-DenseNet, a hybrid model of SpikeGoogle and DenseNet. Experimental outcomes reveal that the proposed method achieved an accuracy of 96.2 %, sensitivity of 97.3 %, and specificity of 94.5 %. [Display omitted] •To detect epileptic seizure, SpikeGoogle-DenseNetis created by interrelating SpikeGoogle with DenseNet.•To select the features, FrAdadelta-CSA is employed by integrating Fractional Calculus into Adadelta-CSA.•To assess effectiveness of suggested schemes, performance metrics like accuracy, sensitivity, and specificity are utilized.