EEG Classification of Epilepsy Based on Machine Learning

The parameter settings of electrical stimulation techniques vary from person to person. Therefore, the transition from a fixed model to an individualized treatment plan is a major challenge in the clinical management of epilepsy. To fully identify the different epileptic states and provide a theoret...

Full description

Saved in:
Bibliographic Details
Published in2024 IEEE 25th China Conference on System Simulation Technology and its Application (CCSSTA) pp. 1 - 6
Main Authors Tang, Xiaomin, Li, Shanshan, Ma, Lerong, Qin, Qing
Format Conference Proceeding
LanguageEnglish
Published IEEE 21.07.2024
Subjects
Accuracy
Bagging
bagging algorithm
Brain modeling
decision tree
Decision trees
Discrete wavelet transforms
Electrical stimulation
electroencephalogram
Electroencephalography
Epilepsy
epilepsy classification
Feature extraction
Measurement
Online AccessGet full text

Cover

More Information
Summary:The parameter settings of electrical stimulation techniques vary from person to person. Therefore, the transition from a fixed model to an individualized treatment plan is a major challenge in the clinical management of epilepsy. To fully identify the different epileptic states and provide a theoretical basis for individualized treatment, a highly robust Bagging classifier based on decision trees was used to classify the epileptic signals into four categories. First, four features such as standard deviation (Std), coefficient of variation (CV), logarithm of amplitude (Log) and interquartile range (IQR) are extracted from the epileptic signals as inputs to the classifier. Then, the optimal feature was selected by feature comparison experiments and the influence of different frequency bands on classification effect was explored by Discrete Wavelet Transformation (DWT). Five metrics, such as confusion matrix, accuracy, correctness, misclassification rate, and F1 score, are provided as evaluation metrics. Finally, the full-band data featuring Std_IQR gave the best classification with accuracy, precision, false positive rate and F1 score of 94.8864%, 94.9904%, 1.7045% and 94.8947%, respectively. Therefore, the method proposed in this paper can classify interictal, preictal, ictal and postictal epilepsy with high-quality.
DOI:10.1109/CCSSTA62096.2024.10691800