Clustering with Minimum Spanning Trees
I compared two Minimum Spanning Tree (MST)-based construction algorithms, Kruskal's and Prim's, to the popular K-Means approach. After benchmarking both on synthetic datasets, Kruskal's performed better and was then tested against K-Means across 200+ real-world datasets using the Adjusted Asymmetric Accuracy (AAA) metric. While K-Means performed more consistently, MST-based clustering showed a clear advantage in detecting non-convex patterns and irregularly shaped clusters, which K-Means struggles with due to its assumption of spherical clusters around centroids.