Cluster Analysis Chart

What it does.

The Cluster Analysis Chart by CBT finds the natural groups in your data and draws them. A scatter plot tells you where points sit; it does not tell you which points belong together. This visual answers that question directly, inside the report, with no data science pipeline in front of it.

Three clustering algorithms are built in, implemented from scratch in TypeScript. Nothing is sent to a service, no R or Python runtime is required, and the clustering recomputes live as slicers and filters change, so the segments always reflect what the user is actually looking at.

Three algorithms. K-Means for compact, evenly sized groups; DBSCAN for arbitrary shapes and explicit noise detection; Hierarchical for nested structure with Ward, Single, Complete, or Average linkage.

Automatic cluster count. Let the visual choose k by the elbow method or by silhouette score across a search range you set, or fix the number of clusters manually.

Dendrogram view. Switch Hierarchical clustering into a dendrogram to see the full merge tree, the cut height, and how the clusters nest. Available vertically or horizontally.

Outlier detection. DBSCAN separates noise points from the clusters and draws them apart, either hollow grey or in a colour you choose, so anomalies are visible rather than absorbed.

Cluster shape. Optional centroid markers, convex hulls, and confidence ellipses convey each cluster's position and spread. Jitter separates points that would otherwise overlap.

Summary table. A pop-down table of per-cluster averages for every variable, shaded against the population average so above and below read at a glance.

Custom cluster names. Rename Cluster 1, Cluster 2 and so on to the business meaning they carry, applied across the legend, tooltips, and summary table.

Standardization. Z-Score, Min-Max, or none, so variables on different scales contribute fairly to the distance calculation rather than the largest-magnitude column dominating.

Variable selection. Choose which bound measures feed the distance maths from a toolbar pop-down, without removing them from the visual or losing them from the tooltips.

Small multiples. Split into a trellis grid to compare clustering across regions, periods, or categories, with optional axis coupling.

Full interactivity. Cross-filtering, drill-down, drill-through, report page tooltips, and bookmark support all work out of the box.

Accessibility. Keyboard navigation, screen-reader labels, high-contrast mode support, and keyboard-only focus indication.

Use Cases: Customer segmentation by spend, frequency, and recency, Product grouping by price and performance characteristics, Store or branch clustering by size and turnover, Anomaly and outlier detection in operational metrics and Sensor or quality readings grouped by behavior