Performance comparison of machine learning techniques in sleep scoring based on wavelet features and neighboring...

Area of Science:

• Computational neuroscience
• Medical informatics

Background:

• Accurate sleep scoring is crucial for diagnosing and treating sleep disorders.
• Manual sleep stage annotation is labor-intensive and requires expertise.
• Machine learning offers an automated solution for sleep scoring.

Purpose of the Study:

• To compare the performance of Support Vector Machines (SVM) and Artificial Neural Networks (ANN) for automated sleep scoring.
• To evaluate the effectiveness of wavelet tree features combined with Neighborhood Component Analysis (NCA) for feature selection and dimension reduction.

Main Methods:

• Utilized the Sleep-EDF polysomnography dataset for analysis.
• Implemented and compared SVM and ANN algorithms.
• Employed wavelet tree features and NCA for feature engineering and selection.

Main Results:

• NCA significantly reduced feature dimensions through its combined linear and non-linear feature selection.
• ANN achieved an accuracy of 90.30% in sleep scoring.
• SVM achieved an accuracy of 89.93% in sleep scoring.

Conclusions:

• The proposed machine learning approach demonstrates acceptable performance for automated sleep scoring, comparable to state-of-the-art methods.
• Integrating NCA with feature generation techniques can further enhance performance.
• Intelligent techniques hold promise for future advancements in diagnosing and treating sleep disorders.