Classification of low-density EEG for epileptic seizures by energy and fractal features based on EMD

EEG Channel-Selection Method for Epileptic-Seizure Classification Based on Multi-Objective Optimization

“…With a set of features extracted, we can train a machine learning model that can reject new instances for prediction in real-time. The works presented in the literature suggest that using methods for feature extraction, is possible to improve the classifier's performance, especially decomposing the EEG signals into different frequency bands, using EMD or DWT (Khan et al, 2012;Sharma and Pachori, 2015;Moctezuma and Molinas, 2019a). The selection of the machine learning method that works better for epileptic seizure classification is also relevant and it has been studied in the literature, however, depending on the feature extraction methods, the classifier's performance may vary.…”

“…The EEG data for each epileptic seizure and epileptic-free period is of 6 s and there are 80 instances on average for each class for each patient. The EEG signals were down-sampled to 128 Hz as our previous research has been shown that the results did not differ using 256 or 128 Hz, however, the process for decomposing the EEG signals into different sub-bands is faster with 128 Hz (Moctezuma and Molinas, 2019a). More details can be found in Goldberger et al (2000), Shoeb (2009), and Moctezuma and Molinas (2019a).…”

“…The method for feature extraction consisted of computing two energy values and two fractal dimension values from both the EMD (Moctezuma andMolinas, 2018, 2020) and DWT subbands (Moctezuma and Molinas, 2019b). This set of features are introduced in Moctezuma and Molinas (2018), Moctezuma and Molinas (2019a), and Moctezuma and Molinas (2020) and described below.…”

“…We use 10-fold cross-validation to evaluate the performance of each classifier. It should be noted that 9fold cross-validation was applied when required, depending on the lowest number of trials per class in the patients, i.e., in the case of subject 16, according to the information described in Goldberger et al (2000), Shoeb (2009), and Moctezuma and Molinas (2019a).…”

“…We previously presented an approach using empirical mode decomposition (EMD) to decompose EEG signals into different intrinsic mode functions (IMF) and five features for each chosen IMF (Moctezuma and Molinas, 2019a). In that study, we presented results of an approach based on channel reduction using the backward-elimination algorithm, obtaining an average classification accuracy of 0.93 when five channels and 10-fold cross-validation were used.…”