RETRACTED ARTICLE: Automatic epileptic seizure recognition using reliefF feature selection and long short term memory classifier
| Autor: | Hirald Dwaraka Praveena, K. Rama Naidu, Chennapalli Subhas |
|---|---|
| Rok vydání: | 2020 |
| Předmět: |
False discovery rate
General Computer Science medicine.diagnostic_test Computer science business.industry Feature vector Feature extraction Feature selection Pattern recognition 02 engineering and technology Electroencephalography Hilbert–Huang transform 03 medical and health sciences 0302 clinical medicine 0202 electrical engineering electronic engineering information engineering medicine 020201 artificial intelligence & image processing Epileptic seizure Artificial intelligence medicine.symptom business Classifier (UML) 030217 neurology & neurosurgery |
| Zdroj: | Journal of Ambient Intelligence and Humanized Computing. 12:6151-6167 |
| ISSN: | 1868-5145 1868-5137 |
| Popis: | Electroencephalogram (EEG) signal based epileptic seizure recognition is an emerging technique that efficiently identifies the non-stationary progresses of brain activities. Usually, the epilepsy is recognized by clinicians on the basis of visual observation of EEG signals that normally consumes more time and also sensitive to noise. In this research, a new supervised system is proposed for automatic epileptic seizure detection. Initially, the signals are collected from Temple University Hospital (TUH), Bern–Barcelona EEG (BB-EEG), Bonn University EEG (BU-EEG) and simulated database. Then, fast empirical mode decomposition (FEMD) and feature extraction (combination of entropy, frequency, auto-regressive, and statistical features) are employed for extracting the features from collected data. Besides, reliefF approach is used to lessen the number of extracted features by obtaining a set of active feature vectors that completely solves the “curse of dimensionality” issue. After feature selection, a supervised classifier [long short-term memory (LSTM)] is utilized for classifying the epileptic seizure classes. Experimental phase demonstrates that the proposed work efficiently classifies the epileptic seizure classes by means of miss rate, specificity, false discovery rate (FDR), false omission rate (FOR), sensitivity, and accuracy. From the experimental result, the proposed work improves the classification accuracy upto 0.6–16% related to the existing works. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|
Full text from SpringerLink