HDBSCAN Clustering on Wine (UCI wine recognition)

HDBSCAN Clustering on Wine (UCI wine recognition) produced 2 clusters and 70 noise rows (silhouette 0.360, adjusted Rand 0.266 against held-out wine_class).

Results of this run

HDBSCAN stability on Wine (UCI wine recognition) extracts clusters at minClusterSize 5. Stability silhouette is 0.360, stability Calinski-Harabasz is 60.8, stability Davies-Bouldin is 1.056. Held-out wine_class versus stable clusters: adjusted Rand 0.266. 70 noise rows out of 178 fell out of every stable branch. Stable cluster 0 on Wine (UCI wine recognition) holds 29 rows (proline 642.2 / color_intensity 7.37). Stable cluster 1 on Wine (UCI wine recognition) holds 79 rows (proline 868.8 / color_intensity 4.47). min_cluster_size sweep: min_cluster_size 5 extracts 0.360 with 2 groups on Wine (UCI wine recognition); min_cluster_size 8 extracts 0.361 with 2 groups on Wine (UCI wine recognition); min_cluster_size 15 extracts 0.351 with 2 groups on Wine (UCI wine recognition). Stability plot keeps 55.4% of variance.

HDBSCAN Clustering on Wine (UCI wine recognition): PCA plot of 2 clusters and 70 noise rows
Each point is one row. Gray points are noise. Computed 2026-11-02 with scikit-learn 1.4.2.
min_cluster_sizeSilhouetteClusters
50.3602
80.3612
150.3512
# Product HDBSCAN: min_cluster_size=5, min_samples=5, metric=euclidean.
# Features are StandardScaler-transformed before the fit.

Who, how, and why

Method: StandardScaler, drop constant columns. Code version 3e4afc4b4352 (scripts/seo/build_explore_artifacts.py). scikit-learn 1.4.2. Seed 42. Dataset source: UCI Machine Learning Repository, wine recognition. License: CC BY 4.0. The wine_class column was held out of the fit and used only for the adjusted Rand score.

Results computed on 2026-11-02.

The results paragraph is filled from these numbers by a fixed template in the site code, so the scores can be checked against the same dataset in the tool.

Maintainer: nikolaou1982.

Contact: nick1982100@gmail.com.

How HDBSCAN Clustering Works

HDBSCAN extends DBSCAN to handle varying-density clusters. It builds a hierarchy and extracts the most stable clusters automatically.

About the Wine (UCI wine recognition) Dataset

178 samples and 13 chemical measurements from three cultivars. The column held out of clustering is the cultivar, not a quality score. Source: UCI Machine Learning Repository, wine recognition. License: CC BY 4.0.

Samples. 178

Features. 13

Type. Numeric

Category. Density-based

When to Use HDBSCAN Clustering

HDBSCAN Clustering belongs to the Density-based family of clustering algorithms. These methods identify clusters as regions of high point density separated by regions of low density. They can discover clusters of arbitrary shape.

Related Examples

Ready to try HDBSCAN Clustering on your own data?

Upload a CSV and get instant results — free to start.