An automated snoring sound classification method based on local dual octal pattern and iterative hybrid feature...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Snoring sound classification (SSC) is crucial for diagnosing sleep disorders.
• Existing methods often struggle with low success rates on datasets like the Munich-Passau Snore Sound Corpus (MPSSC).
• There is a need for more accurate and robust SSC techniques.

Purpose of the Study:

• To introduce a novel snoring sound classification (SSC) method.
• To enhance classification accuracy and success rates for snoring sounds.
• To address limitations of current SSC approaches.

Main Methods:

• Proposed a new feature extractor: Local Dual Octal Pattern (LDOP).
• Integrated multilevel discrete wavelet transform (DWT) decomposition with LDOP for feature generation.
• Employed ReliefF and iterative neighborhood component analysis (RFINCA) for feature selection.
• Utilized k-nearest neighbors (kNN) with leave-one-out cross-validation (LOOCV) for classification.

Main Results:

• The LDOP-based SSC method achieved 95.53% classification accuracy.
• An unweighted average recall (UAR) of 94.65% was obtained.
• The proposed method demonstrated a 22% improvement over state-of-the-art methods.

Conclusions:

• The novel LDOP-based SSC method is highly effective.
• The combination of DWT, LDOP, RFINCA, and kNN yields superior performance.
• This research offers a significant advancement in snoring sound classification technology.