Machine Learning-Based Models for the Pre-Emptive Diagnosis of Epileptic Seizure Using Clinical Data

Chronic diseases are a major burden on the world’s healthcare systems since they require ongoing care and harm the quality of life of those who are afflicted. Among these, epilepsy is most notable as a common neurological condition with repeated seizures that affects more than 50 million individuals globally. Epilepsy diagnosis can be challenging, especially in healthcare settings with limited resources. Nevertheless, early detection of epileptic seizures is essential for optimal management. The paper investigates the use of machine learning algorithms for the preemptive diagnosis of epileptic seizures by utilizing diagnostic, clinical, and demographic data. To distinguish between seizures that are epileptic and those that are not, three machine learning methods are used: Gradient Boosting, Support Vector Machine, and Logistic Regression. The dataset from patients with epileptic seizures (ES) or patients with non-epileptic seizures (NES) was gathered between January 1, 2017, and May 15, 2019. This study optimizes the model using 10-fold GridSearchCV. The use of SelectKBest to decrease the number of features and improve predicted accuracy is a key component of the process. The results showed that Logistic Regression outpaced other methodologies, using 36 features, where it achieved an accuracy of 87.67% with precision, recall, and F1 values all of 88%.