1. Automated detection of epileptic EEG signals using recurrence plots-based feature extraction with transfer learning.
- Author
-
Goel, Sachin, Agrawal, Rajeev, and Bharti, R. K.
- Subjects
DEEP learning ,FEATURE extraction ,FEATURE selection ,ELECTROENCEPHALOGRAPHY ,SIGNAL processing ,EPILEPSY - Abstract
"Epilepsy" is a common neurological brain disorder that may affect the human being at any stage of life. Electroencephalogram (EEG) signal is the most important tool for the early detection of epileptic seizures in several applications of epilepsy diagnosis. Recent research has carried out various possibilities of predicting and analyzing epileptic seizures by using conventional methods which employed EEG signal processing for feature extraction and further classified by either one of deep learning/ machine learning-based methods. So, there is a requirement to find a suitable and more reliable method to detect and classify EEG signals as epileptic and non epileptic. This paper proposes a non-conventional method in which EEG time series signals are transformed into multiple recurrence plots. Transfer learning methods are used to extract features from these recurrence plots and relevant features are selected by using principal component analysis thereby reducing the computational complexity. These features are fed into various machine learning classifiers which include Decision Tree, K-Nearest Neighbor, Gaussian, Random Forest, Bagging, and Support Vector Machine. Among these classifiers, SVM gives the best performance with 98.15% sensitivity, 99.61% precision, and 98.21% accuracy. This proves to be a novel approach as it evaluates the performance of different machine learning classifiers in two ways. First, with the extracted features only and second by using feature selection on extracted features. The outcomes confirm that employing feature selection enhances the performance of each classifier, thereby validating the effectiveness of feature extraction using deep learning. The proposed work has the potential to classify the epileptic and non-epileptic EEG signal more efficiently in terms of tradeoffs between sensitivity and accuracy. This is an added advantage as it uses the non-conventional and hybrid approach to classify the raw EEG signal and thus opening the possibility of intervention required to prevent the possibility of upcoming seizure. [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF