customer segmentation

Segmentation turns a transaction log into named buyer groups. The canonical practice dataset is UCI Wholesale Customers (repository id 292), donated by Margarida Cardoso in March 2014: 440 rows x 8 columns covering annual spend in monetary units on Fresh, Milk, Grocery, Frozen, Detergents_Paper and Delicassen, plus two label columns - Channel (298 Horeca versus 142 Retail) and Region (Lisbon 77, Oporto 47, Other 316).

Those labels make it self-checking: PCA, k-means and hierarchical clustering exercises routinely recover the channel split from Detergents_Paper-versus-Fresh/Frozen spending patterns. Descriptive statistics ship with the data - Fresh ranges 3 to 112,151 with mean around 12,000 - and there are no missing values. It distributes as a 14.7 KB CSV under commercial delivery terms 4.0 with a required citation.

Segmentation quality decides where margin leaks. Mislabelled segments misdirect promotions, credit limits and stock allocation, and bad input data produces confident-looking clusters that do not survive contact with next quarter.

• Labels are ground truth for validation. The UCI benchmark's 298 Horeca versus 142 Retail labels let you score unsupervised clusters against reality - demand equivalent labels from any vendor.
• Category choice shapes clusters. Six spend dimensions yield the classic channel separation; fewer or coarser categories may not, so audit feature coverage before trusting segments.
• Static snapshots age quickly. This 440-row file is a single annual snapshot from one Portuguese distributor - useful for method development, not for current market sizing.

1. Confirm one-row-per-customer grain. UCI documents exactly that - 440 customers x 8 columns; ask other sources to state their grain equally precisely.

1. Demand labeled channels or regions. Channel (Horeca vs Retail) and Region (Lisbon, Oporto, Other) columns make validation possible; unlabeled spend matrices cannot be scored.

1. Check category completeness. Six product categories spanning fresh food through detergents and delicatessen give clustering enough spread to separate behaviour.

1. Verify licensing and citation terms. commercial delivery terms 4.0 with required citation Cardoso (2014/2013), DOI 10.24432/C5030X - free for any purpose with credit.

1. Prefer programmatic access for teaching pipelines. fetch_ucirepo(id=292) pulls it directly into Python alongside the 14.7 KB CSV download.

Entries that sit next to customer segmentation in this glossary:

• wholesale-trade