The Development of an Information Criterion for Change-Point Analysis

Area of Science:

• Statistics
• Time Series Analysis
• Data Science

Background:

• Change-point analysis identifies transitions in systems with discrete states.
• Existing methods include frequentist and information-based approaches, which are often disparate.
• Time series data corrupted by noise presents challenges for accurate state transition detection.

Purpose of the Study:

• To present a unified information-based approach for change-point analysis.
• To reconcile frequentist and information-based methods for testing state transitions.
• To develop a statistically principled, parameter-free method for change-point problems.

Main Methods:

• Developed a unified information-based framework for change-point testing.
• Reconciled previously separate frequentist and information-based change-point detection strategies.
• Applied the method to time series data with discrete state transitions and noise.

Main Results:

• The unified approach effectively tests for the existence of change points.
• The method is statistically principled and does not require prior information or parameter setting.
• Demonstrated wide applicability across various change-point analysis scenarios.

Conclusions:

• The proposed unified method offers a flexible and computationally tractable solution for change-point analysis.
• This approach provides a robust and widely applicable tool for identifying state transitions in time series.
• The parameter-free nature simplifies application and enhances the reliability of change-point detection.