DABiG: Breath pattern classification using the hybrid deep learning with optimal feature selection.

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Machine Learning for Health

Background:

• Breathing patterns reflect emotional and physical well-being.
• Analysis of breathing patterns offers insights into health status.

Purpose of the Study:

• To classify diverse breathing patterns using sensor data.
• To develop a comprehensive approach for breathing pattern recognition.

Main Methods:

• Utilized gyroscope and accelerometer data from two sensors for six breathing patterns.
• Applied Min-Max normalization for data pre-processing.
• Introduced Adaptive Chimp Optimization (AdCO) for feature selection.
• Developed a hybrid deep learning model, DABiG, employing Bidirectional Gated Recurrent Units (BiGRU).

Main Results:

• AdCO demonstrated enhanced convergence and global optimal feature selection.
• The DABiG model incorporated spatial and temporal attention mechanisms.
• Attention mechanisms improved focus on relevant data regions and time steps.

Conclusions:

• Spatial and temporal attention mechanisms significantly enhance feature extraction.
• The proposed method improves the accuracy of breathing pattern classification.
• This approach offers a robust tool for analyzing physiological and emotional states through breathing.