# A review of change point detection methods

**Authors:** Charles Truong, Laurent Oudre, Nicolas Vayatis  
**arXiv:** [1801.00718](https://arxiv.org/abs/1801.00718)  
**Date:** 2018-01-02  
**Summary page:** https://www.summarizepaper.com/en/paper/1801.00718v1/

## Key points

- The paper provides a comprehensive overview of change point detection methods in multivariate time series.
- Retrospective procedures such as maximum likelihood estimation, regression, and kernel methods are discussed.
- Three key elements for change point detection are introduced: a cost function, a search method, and a constraint on the number of changes.
- Each method is described in detail including signal model, algorithm, theoretical guarantees, and application domain.
- The framework aims to facilitate the prototyping of change point detection methods for specific segmentation tasks.
- The paper highlights the significance of this work in addressing challenges in change point detection.
- It offers valuable insights for researchers and enables better understanding and comparison of techniques.
- The comprehensive approach promotes further advancements in change point detection methodologies.

## Summary

The paper titled "A review of change point detection methods" provides a comprehensive overview of various techniques used to detect one or multiple change points in multivariate time series. The authors discuss retrospective (off-line) procedures such as maximum likelihood estimation, regression, and kernel methods, emphasizing that this field of research encompasses numerous applications and diverse models with different operational constraints. To provide a formal framework for change point detection, the authors introduce three key elements: a cost function, a search method, and a constraint on the number of changes to detect. Each method is described in detail including the assumed signal model, associated algorithm, any theoretical guarantees and application domain. This framework aims to facilitate the prototyping of change point detection methods by allowing researchers to appropriately select these elements when designing an algorithm for a specific segmentation task. The paper highlights the significance of this work in addressing the challenges associated with change point detection in multivariate time series. By reviewing various methods and providing a formal framework it offers valuable insights for researchers working in this field. The authors' comprehensive approach enables better understanding and comparison of different techniques while also promoting further advancements in change point detection methodologies.