A breakpoint detection error function for segmentation model selection and evaluation

TL;DR

This paper introduces the breakpointError function for evaluating and selecting segmentation models by accurately measuring the discrepancy between estimated and true breakpoints, with practical implementation in R.

Contribution

It presents a novel error function for breakpoint detection evaluation and demonstrates its application for optimal penalty selection in segmentation models.

Findings

BreakpointError effectively evaluates breakpoint detection accuracy.

The function can be relaxed to handle real data annotations.

A fast C implementation is provided in an R package.

Abstract

We consider the multiple breakpoint detection problem, which is concerned with detecting the locations of several distinct changes in a one-dimensional noisy data series. We propose the breakpointError, a function that can be used to evaluate estimated breakpoint locations, given the known locations of true breakpoints. We discuss an application of the breakpointError for finding optimal penalties for breakpoint detection in simulated data. Finally, we show how to relax the breakpointError to obtain an annotation error function which can be used more readily in practice on real data. A fast C implementation of an algorithm that computes the breakpointError is available in an R package on R-Forge.