EEG-based epilepsy detection using CNN-SVM and DNN-SVM with feature dimensionality reduction by PCA.

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Machine Learning in Healthcare

Background:

• Epilepsy detection from electroencephalogram (EEG) signals is crucial for diagnosis and treatment.
• Traditional machine learning methods often require extensive feature engineering.
• Deep learning models offer potential for automated feature extraction but can be computationally intensive.

Purpose of the Study:

• To evaluate hybrid deep learning and machine learning models for accurate epilepsy detection.
• To assess the impact of feature dimensionality reduction using Principal Component Analysis (PCA).
• To compare the performance of Convolutional Neural Network-Support Vector Machine (CNN-SVM) and Deep Neural Network-Support Vector Machine (DNN-SVM) architectures.

Main Methods:

• Implementation of hybrid CNN-SVM and DNN-SVM models.
• Application of Principal Component Analysis (PCA) for feature dimensionality reduction.
• Evaluation of models on two benchmark EEG datasets: Epileptic Seizure Recognition and BONN.

Main Results:

• The CNN-SVM-PCA model achieved high accuracy: 99.42% on the Epileptic Seizure Recognition dataset and 99.96% on the BONN dataset.
• The DNN-SVM model with PCA also showed improved accuracy, particularly on the BONN dataset (+3.07%).
• Individual CNN and DNN models demonstrated strong baseline performance before PCA and SVM integration.

Conclusions:

• Hybrid CNN-SVM and DNN-SVM models, enhanced by PCA, are highly effective for robust epilepsy detection.
• Feature dimensionality reduction plays a significant role in optimizing model performance and generalization.
• These advanced computational approaches show promise for improving clinical diagnosis of epilepsy.