# Stop using the elbow criterion for k-means and how to choose the number of clusters instead

**Authors:** Erich Schubert  
**arXiv:** [2212.12189](https://arxiv.org/abs/2212.12189)  
**Date:** 2022-12-23  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2212.12189v1/

## Key points

- The elbow criterion is commonly used to determine the optimal number of clusters in k-means clustering
- Relying on the elbow method can lead to poor conclusions
- Better alternatives for determining cluster numbers have been available in literature for a long time
- The authors advocate for completely abandoning the elbow method due to its lack of theoretical support
- Educators should discuss the limitations of the elbow method and teach students about alternative approaches
- Researchers and reviewers should reject any conclusions drawn from the elbow method
- This letter serves as a call-to-action for the academic community to move away from relying on the elbow criterion
- Alternative methods that offer more reliable results and possess stronger theoretical foundations should be explored and adopted.

## Summary

The letter titled "Stop using the elbow criterion for k-means and how to choose the number of clusters instead" addresses a major challenge in k-means clustering, which is determining the optimal number of clusters (k). The authors emphasize that relying on the commonly used heuristic known as the "elbow method" can lead to poor conclusions. They highlight that better alternatives have been available in literature for a considerable time and aim to draw attention to these alternative methods, which often yield superior results. The authors strongly advocate for completely abandoning the elbow method due to its lack of theoretical support. They urge educators to discuss the limitations of this method if it is introduced in class at all and instead teach students about alternative approaches. Furthermore, they encourage researchers and reviewers to reject any conclusions drawn from the elbow method. This letter serves as a call-to-action for the academic community to move away from relying on the elbow criterion for determining cluster numbers in k-means clustering. It emphasizes the need for exploring and adopting alternative methods that offer more reliable results and possess stronger theoretical foundations. In conclusion, this letter encourages academics to abandon outdated techniques such as the elbow method in favor of more reliable approaches with stronger theoretical backing.