UCI Machine Learning Repository review: the best classic AI datasets for learning and benchmarks
Answer in brief: UC Irvine's repository is the classic home of machine learning benchmarks. It maintains 690 datasets, including Iris (150 instances), Heart Disease, Wine Quality and Bank Marketing, and it is still adding new ones in 2026, from wearable-sensor activity data to a vision-language benchmark. It is free and well cited, and perfect for teaching and tabular experiments. It is not a source of foundation-model training data.
Best for: teaching, classical ML benchmarks and small, well-described tabular, sensor and audio datasets.
Independent research from the official UCI repository homepage, checked 2026-11-09. Not hands-on testing.
What it does for AI datasets
Each dataset page lists its task type (classification, regression, clustering), number of instances and number of features. You can browse the most popular and newest datasets, and anyone can contribute one. The homepage also invites visitors to try a new version of the site.
The datasets worth grabbing here, by task
- Classic classification and regression: Iris (150 instances, 4 features), Heart Disease (303), Wine Quality (4.9K), Bank Marketing (45.21K), Breast Cancer Wisconsin (569), Student Performance (649).
- Vision-language evaluation: COREVQA, 5,608 images paired with synthetic true/false statements to test visual entailment in crowded scenes.
- Sensors and time series: human activity recognition from a 9-axis accelerometer (30 adults, IMU signals at 100 Hz).
- Audio: HLS-CMDS, 535 heart and lung sound recordings from a clinical manikin.
- Larger tabular: Amazon Product and Google Locations Reviews, with 3.06M instances of hourly review counts.
Who it's built for
Teachers, students, authors of methods papers and anyone who needs a well-known baseline dataset that reviewers will recognize.
Where it shines
- Recognized everywhere: its classics show up in coursework and papers.
- Clear metadata: task, instance and feature counts at a glance.
- Still growing: new datasets keep arriving, including AI-focused ones.
Honest drawbacks
- The datasets are small by modern AI standards.
- There is nothing here for LLM pretraining, code or robotics.
- Licenses and documentation depth vary by dataset, so check each page.
Pricing and licenses
Free. Licenses are listed per dataset.
Best alternatives
- OpenML: consistently formatted datasets with benchmarking APIs.
- Kaggle: applied datasets with free notebooks.
FAQ
How many datasets does UCI have?
The homepage says it currently maintains 690 datasets.
Is the UCI repository free?
Yes. You can find, download and contribute datasets for free.
Can I train a large language model on UCI data?
No. UCI datasets are small benchmark datasets. For pretraining, see FineWeb or Dolma.
Sources
- UCI Machine Learning Repository
- Iris dataset page
- COREVQA dataset page
← Back to the ranked list of the best AI datasets
Named on this page? Join the network for a linked profile, Visit buttons and featured placement.