Abstract:
Anomaly detection in electroencephalogram (EEG) seizure recognition is crucial for timely and effective patient intervention. Traditional machine learning techniques ofte...Show MoreMetadata
Abstract:
Anomaly detection in electroencephalogram (EEG) seizure recognition is crucial for timely and effective patient intervention. Traditional machine learning techniques often depend on well-labeled data and defined anomaly patterns, which are not always feasible or available in practical situations. This research investigates the application of autoencoders, a form of advanced learning model, to detect irregularities in EEG data, and compares its performance against a traditional classification algorithm and the Isolation Forest model. We utilize the publicly available EEG seizure recognition dataset from the UCI repository for our experiments. The autoencoder is designed to understand a concise version of normal EEG readings and identify anomalies based on the reconstruction error. The evaluation is based on the AUC-ROC score, a commonly used performance measure. Our findings show that the autoencoder performs much better than the traditional classification algorithm and the Isolation Forest model based on the AUC-ROC score, indicating its superior ability to differentiate between regular and unusual EEG readings. This indicates a hopeful potential of autoencoders in EEG seizure recognition and other anomaly detection tasks, especially in scenarios with limited labeled data or undefined anomaly patterns.
Date of Conference: 14-15 November 2023
Date Added to IEEE Xplore: 25 December 2023
ISBN Information: