Nonparametric segmentation of nonstationary time series

Area of Science:

• Complex systems analysis
• Time series analysis
• Statistical modeling

Background:

• Complex systems often exhibit nonstationary behavior, characterized by periods of quasistationarity.
• Standard data analysis methods typically assume stationarity, necessitating methods to identify stationary intervals.
• Existing methods often rely on low-order statistical moments, potentially missing nuanced changes.

Purpose of the Study:

• To develop and present a novel segmentation algorithm for detecting quasistationary spells in time series.
• To provide a robust method for identifying intervals of stationarity within complex, nonstationary systems.
• To generalize and improve upon existing time series segmentation techniques.

Main Methods:

• A fully nonparametric approach to time series segmentation.
• Algorithm designed for real-time detection of quasistationary intervals.
• Application to diverse real-world time series with varying degrees of nonstationarity.

Main Results:

• The proposed algorithm effectively segments time series into quasistationary spells.
• Demonstrated applicability across various real-world datasets with differing nonstationarity.
• The nonparametric method uncovers features missed by methods relying solely on low-order statistics.

Conclusions:

• The developed segmentation algorithm offers a powerful tool for analyzing nonstationary time series.
• This approach enhances the understanding of complex systems by accurately identifying stable periods.
• The method provides a more comprehensive analysis than traditional statistical moment-based techniques.