Probabilistic Epileptic Seizure Onset Risk Assessment Without Prior Symptoms and Seizure Type Classification

Epileptic seizures pose significant risk to individuals often occurring without prior symptoms. Accurate prediction and estimation of risk of epileptic seizures is a critical challenge faced in neurology. In our study, hybrid model based on CNN & LSTM was employed as it efficiently captures both temporal and spatial dependencies. The model achieves an accuracy of 86% with a mean absolute error of 0.01. Besides classification, unsupervised techniques such as K-Means clustering were used without labeled data for seizure-type classification. Feature importance using an analyzed Random Forest classifier reveals principal features across the time and frequency domains. By demonstrating real time seizure risk prediction and classification of seizure type with practical considerations of deployment, this study bridges the gaps between probabilistic AI and clinical applications and offers calibrated uncertainty metrics, enabling improved trustworthiness in critical clinical scenarios.