Changepoint Detection in Noisy Data Using a Novel Residuals Permutation-Based Method (RESPERM): Benchmarking and...

Area of Science:

• Neuroscience
• Medicine
• Time Series Analysis

Background:

• Detecting model changes in noisy time series is crucial for fields like psychophysiology and treatment monitoring.
• Existing methods may struggle with high noise levels common in single trial data.

Purpose of the Study:

• Introduce a novel method, the residuals permutation-based method (RESPERM), for single changepoint detection in linear time series regression.
• Compare RESPERM's performance against the established SEGMENTED method.

Main Methods:

• RESPERM identifies the optimal changepoint by maximizing Cohen's effect size, using parameters from permuted residuals in a linear model.
• Extensive simulations were conducted, varying noise levels, distributions, and changepoint locations.
• The RESPERM method was compared against the SEGMENTED method.

Main Results:

• RESPERM demonstrated consistently lower variance in detected changepoints compared to the SEGMENTED method, particularly in time series with medium to large amounts of noise.
• The methods were applied to a real-world dataset of N250 ERP component amplitudes during face learning.

Conclusions:

• The residuals permutation-based method (RESPERM) is well-suited for changepoint detection, especially in noisy datasets.
• RESPERM offers a robust alternative to existing methods, particularly valuable in neuroscience and medical research.