The Shape Metric for Clustering Algorithms

TL;DR

This paper introduces a shape metric to evaluate how accurately clustering algorithms identify the shape of clusters, providing a new way to assess and improve clustering performance.

Contribution

The paper proposes a novel shape metric for clustering evaluation and applies it to well-known algorithms, suggesting improvements for future methods.

Findings

Shape metric effectively measures cluster shape recognition.

Existing algorithms vary significantly in shape detection accuracy.

Recommendations for enhancing clustering algorithms' shape identification.

Abstract

We construct a method by which we can calculate the precision with which an algorithm identifies the shape of a cluster. We present our results for several well known clustering algorithms and suggest ways to improve performance for newer algorithms.