Advanced sleep disorder detection using multi-layered ensemble learning and advanced data balancing techniques.

Area of Science:

• Medical Diagnostics
• Artificial Intelligence
• Machine Learning

Background:

• Sleep disorder diagnosis is crucial but often labor-intensive.
• Machine learning has improved sleep disorder detection accuracy.
• Existing methods face challenges with data imbalance and interpretability.

Purpose of the Study:

• To develop a novel coordination model for enhanced sleep disorder detection accuracy and reliability.
• To improve diagnostic precision using a multi-model ensemble approach.
• To address data imbalance issues common in sleep disorder datasets.

Main Methods:

• Implemented a multi-layered ensemble model with N selected models.
• Utilized thresholding, predictive scoring, and Softmax label conversion for interpretability.
• Employed voting and stacking ensemble techniques for collaborative decision-making.
• Evaluated performance on original and SMOTE-modified datasets to handle data imbalance.

Main Results:

• The ensemble model achieved 96.88% accuracy on the SMOTE dataset and 95.75% on the original dataset.
• Eight-fold cross-validation demonstrated 99.5% accuracy, highlighting reliability with unbalanced data.
• The ensemble method significantly outperformed individual traditional models in accuracy and handling imbalanced data.

Conclusions:

• The proposed ensemble model offers a significant advancement in sleep disorder detection.
• The integration of multiple models and interpretability methods enhances diagnostic accuracy.
• This approach promises improved patient outcomes and broader applications in medical diagnostics.