Epileptic Seizure Prediction: A Comparative Analysis of LSTM and ANN Models
Shaan Manchanda, Akanksha Gupta · 2025
Epileptic seizures, characterized by abnormal electrical discharges in the brain, pose significant challenges to individuals' health and quality of life. The accurate and timely detection of seizures is critical for effective treatment and prevention of severe consequences. Electroencephalogram (EEG) signals, widely used for seizure monitoring, offer valuable insights into neural activity but require robust analysis techniques to identify seizures accurately. Leveraging advancements in machine learning and deep learning, this study investigates the performance of Artificial Neural Networks (ANN) and Long Short-Term Memory (LSTM) networks in classifying seizure and non-seizure activity using the UCI Epileptic Seizure Recognition dataset. The dataset was preprocessed through normalization, outlier handling, and feature scaling, ensuring data quality for effective model training and evaluation. Key metrics, including accuracy, precision, recall, and ROC-AUC scores, were employed to assess the models' performance comprehensively. The findings highlight that the ANN model outperforms the LSTM in terms of classification accuracy and computational efficiency, achieving an impressive ROC-AUC of 0.987. Its simpler architecture effectively captured the dataset's numerical features, leading to fewer misclassifications. In contrast, the LSTM model, with an AUC of 0.900, demonstrated limitations in handling the segmented EEG data, resulting in higher misclassification rates, particularly for seizure instances. However, its ability to model temporal dependencies makes it a promising approach for continuous EEG data analysis. This study underscores the importance of aligning model selection with dataset characteristics, emphasizing that while ANN offers robust performance for segmented EEG data, LSTM holds potential for tasks requiring sequential pattern recognition. The results contribute to advancing machine learning applications in epilepsy diagnosis and treatment.