Outlier Handling Strategy of Ensembled-Based Sequential Convolutional Neural Networks for Sleep Stage Classification.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Sleep Medicine

Background:

• Automatic sleep stage classification is crucial for understanding sleep quality.
• Current methods struggle with individual variations, leading to outlier data and biased models.
• Outliers can negatively impact feature selection and overall model performance.

Purpose of the Study:

• To develop a robust method for automatic sleep stage classification that addresses individual variability.
• To minimize the impact of outlier data on model training and performance.
• To enhance the accuracy and reliability of sleep quality monitoring for individuals.

Main Methods:

• Proposed an ensemble-based sequential convolutional neural network (E-SCNN) integrating machine learning and deep learning.
• Incorporated a clustering module to categorize individuals based on feature distribution and assign personalized weights.
• Utilized convolutional neural networks for robust feature extraction combined with tailored weights.

Main Results:

• Achieved high accuracies: 84.8% on the Sleep-EDF Expanded dataset and 85.5% on the MASS dataset.
• Demonstrated the E-SCNN's effectiveness in alleviating the outlier problem in sleep stage classification.
• Showcased improved model robustness at the individual level.

Conclusions:

• The E-SCNN model effectively reduces bias caused by outliers, leading to more accurate individual sleep stage classification.
• This approach significantly enhances the robustness of sleep quality monitoring systems for personalized use.
• The findings highlight the potential of ensemble methods with clustering for improving AI in healthcare applications.