A Selective Review on Information Criteria in Multiple Change Point Detection

Area of Science:

• Statistics
• Data Science
• Time Series Analysis

Background:

• Change point detection is crucial for understanding dynamic data streams.
• Identifying multiple change points in noisy data presents significant challenges.
• Existing methods, like Bayesian Information Criterion (BIC), show limitations in finite samples.

Purpose of the Study:

• To review and evaluate information criterion-based methods for multiple change point detection.
• To investigate the practical performance of various criteria, including AIC, BIC, and MDL.
• To assess performance under potential model mis-specification and in real-world applications.

Main Methods:

• Comprehensive review of information criterion-based multiple change point detection methods.
• Simulation studies to compare the performance of different criteria (AIC, BIC, MDL variants).
• Case study analysis using SCADA (Supervisory Control and Data Acquisition) signals from wind turbines.

Main Results:

• Performance of information criteria varies significantly in practical scenarios, especially with noisy data.
• Model mis-specification can notably impact the effectiveness of change point detection methods.
• The study highlights the varying efficacy of AIC, BIC, and MDL in real-world data.

Conclusions:

• No single information criterion universally outperforms others for multiple change point detection.
• Practical performance is highly dependent on data characteristics and potential model misspecification.
• Further research is needed to address challenges in developing robust change point detection techniques.