Electroencephalogram Data Analysed Through the Lens of Machine Learning to Detect Signs of Epilepsy

As most epileptic seizures only happen infrequently the EEG is essential for the categorization and diagnosis of epilepsy. Empirical interpretation of a first EEG has a low sensitivity (between 29 and 55% [29–55 percent]) when applied to adult patients. During EEG epochs in which seizures are not present, valuable EEG data is buried among the signals and is thus inaccessible to any clinician who does not have specialized training in EEG analysis. In order to distinguish people with generalized epilepsy from the general population, we create a multi-variate technique employing EEG data to comprehend the functional connectivity of the brain at the sensor level. The temporal, periodic, and time-frequency domains were probed using five different measurements to investigate eight distinct connection properties. The K-Nearest Neighbor method was used to evaluate the solution, and the results were compared to two groups: the first consisted of patients with epilepsy, and the second consisted of patients who had nonepileptic episodes. Both EG and HC groups demonstrated high levels of classification accuracy (89%), although EG showed significant spatial-temporal abnormalities in the front central regions at the beta frequency band compared to HC. Classification accuracy for EG and NEAD was only approximately 79% because of the well-documented comorbidity of NEAD and epileptic episodes. This study suggests that seizure-free EEG data may be used to reliably differentiate between those with HC and those with specialized epilepsy. Although more research is needed to develop a diagnostic tool that might aid in treatment.