Similarity-Based Adaptive Window for Improving Classification of Epileptic Seizures with Imbalance EEG Data Stream

Area of Science:

• Data Science
• Machine Learning
• Medical Informatics

Background:

• Data stream mining is crucial for real-time medical data classification.
• Imbalanced data, where minority classes are underrepresented, poses a significant challenge, biasing traditional classification techniques.
• Existing methods often overcompensate, excessively favoring the minority class.

Purpose of the Study:

• To propose a novel method for balancing data streams in medical data classification.
• To address the bias towards the majority class in imbalanced datasets.
• To preserve the original data distribution as much as possible while improving minority class representation.

Main Methods:

• A new data stream balancing technique is introduced.
• The method employs similarity analysis to select relevant instances from previous data windows.
• Selected minority class instances are integrated into the current data window.

Main Results:

• The proposed method demonstrated promising results on the Siena dataset.
• Performance was compared against the Skew ensemble method and other existing approaches.
• The technique effectively balanced the minority class presence while minimizing disruption to the majority class.

Conclusions:

• The developed method offers an effective solution for imbalanced medical data classification in data streams.
• It provides a balanced approach, improving minority class recognition without significantly altering the majority class distribution.
• This technique holds potential for enhancing the reliability of machine learning models in critical medical applications.