What is the Wholesale Customers Dataset dataset?

440 wholesale customer records from a Portuguese distributor with annual spending across 6 product categories (Fresh, Milk, Grocery, Frozen, Detergents_Paper, Delicassen). A benchmark dataset for customer segmentation, K-means clustering, and market basket analysis.

How can I download the Wholesale Customers Dataset dataset?

You can download the Wholesale Customers Dataset dataset in CSV or Parquet format directly from this page. Each table has its own download buttons.

What format is the Wholesale Customers Dataset dataset available in?

This dataset is available in CSV and Parquet formats. CSV is great for spreadsheet applications, while Parquet is optimized for data analysis tools like Pandas and DuckDB.

What is the license for the Wholesale Customers Dataset dataset?

This dataset is available under the CC BY 4.0 license. See the full license at https://creativecommons.org/licenses/by/4.0/

How many tables are in the Wholesale Customers Dataset dataset?

The Wholesale Customers Dataset dataset contains 1 table: customers.

Wholesale Customers Dataset

Last updated 1 weeks ago•January 2, 2026

Source:UCI Machine Learning Repository

License:CC BY 4.0

Version:1.0

Time:2011

Location:Portugal (Lisbon, Oporto, and other regions)

Created by Dataset Agent

Overview

The Wholesale Customers Dataset captures real purchasing behavior from a wholesale distributor operating in Portugal, originally compiled by Nuno Abreu for his 2011 doctoral thesis and later published through the UCI Machine Learning Repository. This dataset has become a foundational benchmark for unsupervised machine learning, particularly customer segmentation and clustering algorithm evaluation.

The dataset contains 440 customer records with 8 variables and zero missing values, making it immediately usable without data cleaning.

View Source

SQL

SELECT COUNT(*) AS total_records, COUNT(Channel) AS channel_non_null, COUNT(Fresh) AS fresh_non_null FROM customers.csv

Data

Total Records	Channel Non Null	Fresh Non Null
440	440	440
1 row

What distinguishes this dataset is its dual nature: it supports both classification tasks (predicting customer channel based on spending) and clustering tasks (discovering natural customer segments). The clear separation between Hotel/Restaurant/Café (Horeca) and Retail customers creates interpretable ground truth for validating unsupervised learning results.

Understanding the Variables

The dataset comprises 2 categorical identifiers and 6 continuous spending features, all representing annual monetary units (interpretable as Euros):

Why Channel Matters: Horeca customers (hotels, restaurants, cafés) purchase for food service operations—they need bulk fresh ingredients and frozen items. Retail customers stock shelves—they need grocery items and household supplies like detergents. This fundamental business difference creates naturally separable clusters.

Data Preview

Sample Customer Records (First 5 Rows)

Channel	Region	Fresh	Milk	Grocery	Frozen	Detergents Paper	Delicassen
2	3	12,669	9,656	7,561	214	2,674	1,338
2	3	7,057	9,810	9,568	1,762	3,293	1,776
2	3	6,353	8,808	7,684	2,405	3,516	7,844
1	3	13,265	1,196	4,221	6,404	507	1,788
2	3	22,615	5,410	7,198	3,915	1,777	5,185
5 rows

View Source

SQL

SELECT * FROM customers.csv LIMIT 5

Data

Channel	Region	Fresh	Milk	Grocery	Frozen	Detergents Paper	Delicassen
2	3	12,669	9,656	7,561	214	2,674	1,338
2	3	7,057	9,810	9,568	1,762	3,293	1,776
2	3	6,353	8,808	7,684	2,405	3,516	7,844
1	3	13,265	1,196	4,221	6,404	507	1,788
2	3	22,615	5,410	7,198	3,915	1,777	5,185
5 rows

Notice how the fourth row (Channel 1 = Horeca) shows higher Fresh (13,265) and Frozen (6,404) spending but much lower Milk (1,196) and Detergents_Paper (507) compared to the Retail customers above it. This pattern is consistent throughout the dataset.

Descriptive Statistics

Summary Statistics for Spending Variables

Variable	Min	Max	Mean	Std Dev	Median
Fresh	3	112,151	12,000	12,647	8,504
Milk	55	73,498	5,796	7,380	3,627
Grocery	3	92,780	7,951	9,503	4,756
Frozen	25	60,869	3,072	4,855	1,526
Detergents_Paper	3	40,827	2,881	4,768	816
Delicassen	3	47,943	1,525	2,820	965
6 rows

View Source

SQL

SELECT ROUND(MIN(Fresh)) AS min_fresh, ROUND(MAX(Fresh)) AS max_fresh, ROUND(AVG(Fresh)) AS mean_fresh, ROUND(STDDEV (Fresh), 0) AS std_fresh, ROUND(MEDIAN (Fresh)) AS median_fresh FROM customers.csv

Data

Variable	Min	Max	Mean	Std Dev	Median
Fresh	3	112,151	12,000	12,647	8,504
Milk	55	73,498	5,796	7,380	3,627
Grocery	3	92,780	7,951	9,503	4,756
Frozen	25	60,869	3,072	4,855	1,526
Detergents_Paper	3	40,827	2,881	4,768	816
Delicassen	3	47,943	1,525	2,820	965
6 rows

Right-Skewed Distributions: All spending variables show means significantly higher than medians, indicating right-skewed distributions with high-spending outliers. Consider log-transformation before clustering to reduce the influence of extreme values.

Customer Distribution Analysis

The dataset includes 298 Horeca customers (67.7%) and 142 Retail customers (32.3%), reflecting the distributor's business focus on food service clients.

View Source

SQL

SELECT Channel, COUNT(*) AS count, ROUND( COUNT(*) * 100.0 / ( SELECT COUNT(*) FROM customers.csv ), 1 ) AS percentage FROM customers.csv GROUP BY Channel

Data

Channel	Count	Percentage
1	298	67.7
2	142	32.3
2 rows

View Source

SQL

SELECT Channel, COUNT(*) AS count FROM customers.csv GROUP BY Channel

Data

Channel	Count
Horeca (Hotels/Restaurants/Cafés)	298
Retail	142
2 rows

Geographically, 77 customers from Lisbon (17.5%), 47 from Oporto (10.7%), and 316 from other Portuguese regions (71.8%).

View Source

SQL

SELECT Region, COUNT(*) AS count FROM customers.csv GROUP BY Region ORDER BY Region

Data

Region	Count
1	77
2	47
3	316
3 rows

View Source

SQL

SELECT Region, COUNT(*) AS count FROM customers.csv GROUP BY Region ORDER BY Region

Data

Region	Customer Count
Lisbon	77
Oporto	47
Other Regions	316
3 rows

Channel Comparison: Horeca vs Retail Spending Patterns

The two customer channels exhibit dramatically different purchasing behaviors, which is why this dataset excels for both classification (supervised) and segmentation (unsupervised) tasks.

View Source

SQL

SELECT Channel, ROUND(AVG(Fresh)) AS avg_fresh, ROUND(AVG(Milk)) AS avg_milk, ROUND(AVG(Grocery)) AS avg_grocery, ROUND(AVG(Frozen)) AS avg_frozen, ROUND(AVG(Detergents_Paper)) AS avg_detergents, ROUND(AVG(Delicassen)) AS avg_delicassen FROM customers.csv GROUP BY Channel

Data

Category	Horeca (Channel 1)	Retail (Channel 2)
Fresh	13,476	8,904
Milk	3,452	10,717
Grocery	3,962	16,323
Frozen	3,748	1,653
Detergents & Paper	791	7,270
Delicassen	1,416	1,753
6 rows

Key Business Insight: Horeca customers spend 51% more on Fresh and 127% more on Frozen items—they're stocking kitchens. Retail customers spend 312% more on Grocery and 819% more on Detergents_Paper—they're stocking shelves for consumers. This pattern makes Channel highly predictable from spending data.

Correlation Analysis for Market Basket Insights

Strong correlations between product categories reveal natural purchasing patterns valuable for cross-selling strategies and understanding customer behavior.

The strongest correlation is between Grocery and Detergents_Paper (r=0.925)—customers who buy groceries also buy household supplies. This reflects retail supermarket purchasing patterns.

View Source

SQL

SELECT ROUND(CORR(Grocery, Detergents_Paper), 3) AS grocery_detergents_corr FROM customers.csv

Data

Grocery Detergents Corr
0.93
1 row

Product Category Correlation Matrix

Category Pair	Correlation (R)	Interpretation
Grocery ↔ Detergents_Paper	0.925	Very Strong - Retail basket pattern
Milk ↔ Grocery	0.728	Strong - Supermarket staples
Milk ↔ Detergents_Paper	0.662	Moderate-Strong - Consumer goods
Fresh ↔ Frozen	0.346	Weak-Moderate - Food service pattern
Fresh ↔ Delicassen	0.205	Weak - Limited association
5 rows

View Source

SQL

SELECT ROUND(CORR(Fresh, Frozen), 3), ROUND(CORR(Milk, Grocery), 3), ROUND(CORR(Milk, Detergents_Paper), 3), ROUND(CORR(Grocery, Detergents_Paper), 3), ROUND(CORR(Fresh, Delicassen), 3) FROM customers.csv

Data

Category Pair	Correlation (R)	Interpretation
Grocery ↔ Detergents_Paper	0.925	Very Strong - Retail basket pattern
Milk ↔ Grocery	0.728	Strong - Supermarket staples
Milk ↔ Detergents_Paper	0.662	Moderate-Strong - Consumer goods
Fresh ↔ Frozen	0.346	Weak-Moderate - Food service pattern
Fresh ↔ Delicassen	0.205	Weak - Limited association
5 rows

The correlation structure reveals two distinct purchasing clusters: (1) a Retail cluster where Milk, Grocery, and Detergents_Paper are highly correlated, and (2) a Horeca cluster where Fresh and Frozen show moderate association. This structure is why K-means clustering naturally discovers interpretable segments.

Total Spending Analysis

Total annual spending across all customers is €14.62 million, with Fresh products accounting for €5.28 million (36.1%) of all purchases.

View Source

SQL

SELECT SUM(Fresh) + SUM(Milk) + SUM(Grocery) + SUM(Frozen) + SUM(Detergents_Paper) + SUM(Delicassen) AS total_all, SUM(Fresh) AS total_fresh FROM customers.csv

Data

Total All	Total Fresh
14,619,500	5,280,131
1 row

View Source

SQL

SELECT SUM(Fresh), SUM(Milk), SUM(Grocery), SUM(Frozen), SUM(Detergents_Paper), SUM(Delicassen) FROM customers.csv

Data

Category	Total Spending (€)	Percentage
Fresh	5,280,131	36.1%
Grocery	3,498,562	23.9%
Milk	2,550,357	17.4%
Frozen	1,351,650	9.2%
Detergents_Paper	1,267,857	8.7%
Delicassen	670,943	4.6%
6 rows

High-Value Customer Identification

The top 10% of customers (44 records) account for €5.89 million (40.3%) of total spending, demonstrating significant customer value concentration.

View Source

SQL

SELECT SUM( Fresh + Milk + Grocery + Frozen + Detergents_Paper + Delicassen ) AS top_10_spending FROM ( SELECT *, Fresh + Milk + Grocery + Frozen + Detergents_Paper + Delicassen AS total FROM customers.csv ORDER BY total DESC LIMIT 44 )

Data

Top 10 Spending
5,890,000
1 row

Top 5 Highest-Spending Customers

Rank	Channel	Region	Total Spend (€)	Top Category
1	Horeca	Other	176,040	Fresh (€112,151)
2	Retail	Other	163,968	Grocery (€92,780)
3	Horeca	Other	154,003	Fresh (€76,237)
4	Retail	Other	140,567	Milk (€73,498)
5	Horeca	Lisbon	128,940	Fresh (€56,082)
5 rows

View Source

SQL

SELECT Channel, Region, Fresh + Milk + Grocery + Frozen + Detergents_Paper + Delicassen AS total_spend, Fresh, Milk, Grocery FROM customers.csv ORDER BY total_spend DESC LIMIT 5

Data

Rank	Channel	Region	Total Spend (€)	Top Category
1	Horeca	Other	176,040	Fresh (€112,151)
2	Retail	Other	163,968	Grocery (€92,780)
3	Horeca	Other	154,003	Fresh (€76,237)
4	Retail	Other	140,567	Milk (€73,498)
5	Horeca	Lisbon	128,940	Fresh (€56,082)
5 rows

Data Preprocessing Considerations

Before applying machine learning algorithms, consider these preprocessing steps based on the data characteristics:

Standardization is essential: Spending ranges vary dramatically (Fresh: 3-112,151 vs Delicassen: 3-47,943). Use StandardScaler before K-means or hierarchical clustering.
Log-transformation recommended: All variables are right-skewed with outliers. Log(x+1) transformation normalizes distributions and reduces outlier influence.
One-hot encode Channel/Region: For algorithms requiring numeric input, convert categorical variables. However, for pure clustering, consider excluding these to discover segments independently.
Outlier strategy: The dataset has legitimate high-spenders (not errors). Consider Winsorization (capping at 95th percentile) rather than removal for robust clustering.

Pro Tip: When evaluating clustering results, use Channel as ground truth validation. If your algorithm naturally separates Horeca from Retail customers, it's capturing meaningful business segments.

Why This Dataset for Clustering?

The Wholesale Customers Dataset has become a benchmark for clustering algorithms because it offers:

Manageable size: 440 records allows quick iteration during algorithm development and parameter tuning
Real-world origin: Actual business data with interpretable features, unlike synthetic datasets
Natural clusters: Channel variable provides ground truth for validating unsupervised results
Appropriate complexity: 6 continuous features create non-trivial clustering challenges without overwhelming dimensionality
Clean data: Zero missing values means immediate usability
Business interpretability: Discovered segments translate directly to marketing strategies

Limitations and Considerations

Geographic Limitation: This dataset represents a single wholesale distributor in Portugal. Purchasing patterns may not generalize to other countries, markets, or time periods. The data was collected circa 2011.

Sample size: 440 records may be insufficient for complex deep learning models or detecting rare customer segments
Temporal snapshot: Annual totals don't capture seasonality, trends, or purchasing frequency
Limited demographics: No customer size, age, or detailed business type information beyond Channel
Currency/inflation: Monetary values from 2011 Portugal may need adjustment for contemporary analysis
Class imbalance: 2:1 ratio of Horeca to Retail may bias some classification algorithms

Comparison to Alternative Datasets

Choose the Wholesale Customers Dataset when you need a clean, well-understood benchmark for comparing clustering algorithms, teaching unsupervised learning concepts, or rapidly prototyping customer segmentation approaches before scaling to larger datasets.

Table Overview

customers

Contains 440 rows and 8 columns. Column types: 8 numeric.

440 rows8 columns

customers

440

rows

columns

Data Preview

Scroll to see more

Channel	Region	Fresh	Milk	Grocery	Frozen	Detergents_Paper	Delicassen
2	3	12,669	9,656	7,561	214	2,674	1,338
2	3	7,057	9,810	9,568	1,762	3,293	1,776
2	3	6,353	8,808	7,684	2,405	3,516	7,844
1	3	13,265	1,196	4,221	6,404	507	1,788
2	3	22,615	5,410	7,198	3,915	1,777	5,185

Row 1

Channel2

Region3

Fresh12,669

Milk9,656

Grocery7,561

+3 more columns

Row 2

Channel2

Region3

Fresh7,057

Milk9,810

Grocery9,568

+3 more columns

Row 3

Channel2

Region3

Fresh6,353

Milk8,808

Grocery7,684

+3 more columns

Showing 5 of 440 rows

Data Profile

440

rows

columns

100%

complete

171.9 KB

estimated size

Column Types

8 Numeric

High-Cardinality Columns

Columns with many unique values (suitable for identifiers or categorical features)

Fresh(433 unique values)
Grocery(430 unique values)
Frozen(426 unique values)
Milk(421 unique values)
Detergents_Paper(417 unique values)
Delicassen(403 unique values)

Data Dictionary

customers

Column	Type	Description	Example
`Channel`	numeric	Numeric value (range: 1 - 2)	2, 2
`Region`	numeric	Numeric value (range: 3 - 3)	3, 3
`Fresh`	numeric	Numeric value (range: 6,353 - 22,615)	12669, 7057
`Milk`	numeric	Numeric value (range: 1,196 - 9,810)	9656, 9810
`Grocery`	numeric	Numeric value (range: 4,221 - 9,568)	7561, 9568
`Frozen`	numeric	Numeric value (range: 214 - 6,404)	214, 1762
`Detergents_Paper`	numeric	Numeric value (range: 507 - 3,516)	2674, 3293
`Delicassen`	numeric	Numeric value (range: 1,338 - 7,844)	1338, 1776