Early Detection of Epilepsy Seizures using Machine Learning Algorithms
Date
2022-05
Authors
Rios, Cristian (Chris)
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
According to the World Health Organization (WHO), one of the most common neurological diseases is epilepsy with over 50 million people globally diagnosed to date. Epilepsy is defined as a chronic noncommunicable disease of the brain that affects the central nervous system and can result in sudden seizures. The early detection of seizures would allow suffering patients an opportunity to better manage their seizures and improve overall quality of life. In this paper, we aim to detect seizures using different machine learning (ML) algorithms using non-invasive EEG data of epilepsy patients through a publicly available database. We evaluated different signal pre-processing methods to allow for extraction of optimal features that include descriptive statistical parameters, frequency-based trend changes, power spectrum density, and entropy. For some classifiers, optimal features were passed through a Principal Component Analysis (PCA) algorithm to minimize the number of features used and to reduce the dimensionality of the ML model. ML models evaluated for seizure detection included Logistic Regression, Decision Tree, SVM, KNN, Naïve Bayes, and Neural Networks. A confusion matrix was generated for optimal models to compare accuracy, precision, and specificity. Results demonstrated the models were promising but the top three models with highest accuracy were DT, SVM, and NN. Given these results, ML models such as SVM or Neural Networks, can be utilized to create seizure detecting systems that are portable and use EEG data to detect the onset of seizures. Even though the highest accuracy observed was approximately 82.5%, the value of machine learning models show merit for solving seizure detection challenges. In future research studies, we plan to evaluate additional open databases to enhance these ML models further and increase accuracy to a minimum level of 95% to be evaluated in real-world applications.
Description
Keywords
Epilepsy, EGG, Machine learning