Bayesian Model Search for Nonstationary Periodic Time Series.

Area of Science:

• Statistics
• Biomedical Engineering
• Signal Processing

Background:

• Nonstationary time series analysis is crucial for understanding dynamic biological processes.
• Identifying changing oscillatory patterns in physiological data presents significant analytical challenges.

Purpose of the Study:

• To develop a novel Bayesian methodology for analyzing nonstationary time series with time-varying oscillatory behavior.
• To simultaneously estimate change-points and evolving periodicities within time series data.

Main Methods:

• A piecewise oscillatory model with unknown periodicities was employed.
• A trans-dimensional Markov chain Monte Carlo algorithm was utilized for simultaneous updates of change-points and periodicities.
• The methodology was validated through two distinct application case studies.

Main Results:

• The proposed Bayesian approach successfully identified time-varying oscillatory patterns.
• Ultradian oscillations in human skin temperature during night rest were accurately detected.
• Instances of sleep apnea were effectively identified in plethysmographic respiratory traces.

Conclusions:

• The novel Bayesian methodology provides a robust framework for analyzing complex, nonstationary oscillatory time series.
• This approach has significant implications for e-Health and sleep research, enabling more precise detection of physiological events.
• The method's ability to handle changing periodicities enhances its applicability to dynamic biological signals.