Segmentation & clustering

Which groups actually exist in this audience?

Segment respondents with an approach suited to numeric, categorical, or mixed survey data:

• Auto Clustering — finds natural groupings in mixed data and can mark outliers as Noise.
• Balanced Clusters — mixed data with more even segment sizes; choose 2–8 clusters.
• K-Means — distance-based groups on numeric profiles and opinion scales.
• K-Medoids — mixed survey variables with a fixed cluster count; robust centres are real respondents.
• HDBSCAN and Balanced HDBSCAN — density-based groups with noise detection, with the latter producing a more even allocation.
• Latent Class Analysis (LCA) — probabilistic segments from categorical, ordinal, boolean, and multi-select indicators.

Choose a solution you can defend.

Select columns, tune the settings or accept recommended defaults, then review separation with Silhouette score, PERMANOVA, and Cluster Quality. LCA compares class counts with AIC, BIC, and entropy, while session history makes it possible to compare configurations before saving a result.

Make the groups intelligible.

Each run shows the features that distinguish a cluster: numeric z-scores against the overall mean and categorical lift. Save the cluster labels as a project column, then profile the groups in tables, charts, filters, and delivery outputs instead of rebuilding them in another tool.

The algorithm is only the beginning of the segmentation.

A usable solution needs more than a set of labels. Researchers need to understand the variables that distinguish each group, inspect whether the segments are large and coherent enough for the study, and decide how they will be used in the wider project. AddMaple keeps that assessment beside the saved segment column, the tables, and the outputs that follow.