Popular Datasets on the UCI Machine Learning Repository

The UCI Machine Learning Repository currently maintains 689 datasets, and a handful of them are used so widely that they have become standard benchmarks for classification and regression. If you want a small dataset to test a classifier, start with Iris (150 instances, 4 features) or Breast Cancer Wisconsin (Diagnostic) (569 instances, 30 features). If you need something larger for a realistic tabular problem, Bank Marketing (45.21K instances, 17 features) or Wine Quality (4.9K instances, 12 features) are better choices. The table below summarizes the five most prominent ones.

Overview of popular datasets

What each dataset is good for

Iris

A small classic dataset from Fisher, 1936, described by the repository as one of the earliest known datasets used for evaluating classification methods. With only 150 instances and 4 features, it runs in seconds and is best for verifying that a pipeline works end to end, not for judging real-world performance.

Heart Disease

This entry actually combines 4 databases: Cleveland, Hungary, Switzerland, and the VA Long Beach. It has 303 instances and 13 features, making it a common choice for binary classification practice where you also want to handle missing values and mixed feature types.

Wine Quality

Two datasets are included, related to red and white vinho verde wine samples from the north of Portugal. The goal is to model wine quality based on physicochemical tests (see Cortez et al., 2009). It supports both classification and regression, with 4.9K instances and 12 features — useful when you want a medium-sized problem with an ordinal target.

Bank Marketing

The data relates to direct marketing campaigns (phone calls) of a Portuguese banking institution. The classification goal is to predict whether the client will subscribe a term deposit (variable y). At 45.21K instances and 17 features, it is the largest of these five and is well suited to practicing imbalanced classification.

Breast Cancer Wisconsin (Diagnostic)

The Diagnostic Wisconsin Breast Cancer Database has 569 instances and 30 features. It is a compact but higher-dimensional classification set, often used to compare feature selection or dimensionality reduction methods.

How to choose

• Learning or debugging a model: Iris — smallest, fastest, cleanest.
• Practicing preprocessing and missing-value handling: Heart Disease — multiple source databases, mixed data.
• Working with an ordinal or regression target: Wine Quality — supports both task types.
• Testing on a larger, imbalanced tabular problem: Bank Marketing — most instances and features here.
• Comparing feature selection methods: Breast Cancer Wisconsin (Diagnostic) — 30 features on only 569 instances.

All five are listed under Popular Datasets on the repository homepage, alongside Student Performance (649 instances, 33 features, classification and regression) and newer additions such as HLS-CMDS, a heart and lung sounds dataset with 535 recordings. You can browse the full catalog or contribute a dataset directly from the site.