Fractional Adadelta Chameleon Swarm Algorithm Enabled DenseNet for Epileptic Seizure Detection

• Published in: 2025 8th International Conference on Trends in Electronics and Informatics (ICOEI) pp. 1314 - 1321
• Main Authors: G, Indu Salini, I, Sowmy, Sreeja, T.K., Nair, Lekshmi V
• Format: Conference Proceeding
• Language: English
• Published: IEEE 24.04.2025
• Subjects: Accuracy; deep learning; EEG; Electroencephalography; Entropy; Epilepsy; Feature extraction; optimization algorithm; Prediction algorithms; seizure detection; Sensitivity; Signal processing algorithms; Time-domain analysis
• DOI: 10.1109/ICOEI65986.2025.11012932

Epilepsy is a chronic neurological disease characterized by repeated seizures that can severely affect the excellence of existence for those living with such disorders. Electroencephalography (EEG) has been a most important tool in evaluation of brain activity as it relates to diagnosis of epilepsy since it provides valuable information about abnormal cerebral wave patterns associated with seizure events. However, the process of interpreting EEG recordings requires a highly trained expert, as detecting epileptic activity amidst normal brain rhythms can be complex. This manual analysis is both time-consuming and mentally exhausting, highlighting the need for timely, accurate diagnostic methods to improve the efficiency and effectiveness of epilepsy management and treatment. In this research, a methodology for detection of epileptic seizures is formulated through utilization of optimized Deep Learning (DL) based classification techniques. Initial stages of EEG signal processing involve applying a noise removal method called Short-Time Fourier Transform (STFT), from which a whole range of non-linear features like Hurst Exponent, Petrosian and Box Counting, as well as statistical attributes such as mean, standard deviation, variance and Lyapunov exponent (LE), have to be extracted, together with time domain attributes like Hjorth parameters, Skewness, Shannon entropy and Sure entropy and also spectral features AM and FM. Finally, feature selection is done using Fractional Adadelta Chameleon Swarm Algorithm (FrAdadelta-CSA), a novel method that combines principles of fractional calculus into Adadelta Chameleon Swarm Algorithm (Adadelta-CSA). Utilizing the selected features, seizure prediction is conducted through the application of Optimized DenseNet, whereby DenseNet is trained utilizing FrAdadelta-CSA. Empirical assessments indicate that proposed methodology accomplished 96.1% of accuracy, 96.3% of sensitivity and 93.7% of specificity.