How to Contribute a Dataset to the UCI Machine Learning Repository

Updated

The UCI Machine Learning Repository accepts dataset donations through the "Contribute a Dataset" link on its homepage. The repository currently maintains 689 datasets and describes itself as a service where "you can donate and find datasets used by millions of people all around the world." If you have a dataset that is well-documented, ethically sourced, and potentially useful to the machine learning community, the contribution path starts at that link. The exact submission form fields, file size limits, and review timeline are not specified on the homepage, so you will need to follow the link to see the current requirements.

What the Repository Looks For

The homepage does not publish a formal acceptance checklist, but the datasets it highlights suggest the qualities that matter:

  • A clear task framing. Every popular dataset is labeled with a task type — Classification, Regression, or both. Iris is described as "one of the earliest known datasets used for evaluating classification methods," and Wine Quality supports "Classification, Regression."
  • A documented origin. Heart Disease is noted as combining "4 databases: Cleveland, Hungary, Switzerland, and the VA Long Beach." Wine Quality cites "Cortez et al., 2009" and links to the source page. Provenance and citation are visible parts of how datasets are presented.
  • A stated goal. Bank Marketing explains that "the classification goal is to predict if the client will subscribe a term deposit (variable y)." The purpose of the data is spelled out.
  • Reasonable scale and structure. Listed datasets range from 150 instances / 4 features (Iris) to 3.06M instances / 2 features (Amazon Product and Google Locations Reviews), so both small classic datasets and large modern ones are accepted.

If your dataset has a defined prediction or modeling task, a traceable source, and a description that explains what the variables mean, it fits the pattern of what the repository hosts.

Preparing Your Dataset

Before submitting, organize your materials so a reviewer and future users can understand them without contacting you:

  1. Data files. Provide the data in a stable, widely readable format (for example CSV or ARFF). The repository's own dataset pages list instance and feature counts, so make sure those numbers are accurate for your files.
  2. A description document. Explain what each feature represents, what the target variable is, how the data was collected, and any known limitations or biases.
  3. Task type. State whether the dataset is intended for classification, regression, clustering, or another task — the repository labels datasets this way.
  4. Citation and licensing. Include how you want the dataset cited and the terms under which it can be redistributed. The Wine Quality entry shows a citation to the originating paper, and the HLS-CMDS entry describes its recording setup in detail; both are examples of the context the repository surfaces.
  5. Ethical and privacy review. If the data involves people, confirm it is de-identified and that you have the right to share it. The homepage does not state a specific policy, so treat this as your responsibility before submitting.

Submitting Through the Contribution Link

The homepage provides a "Contribute a Dataset" link alongside "View Datasets." The general flow is:

  1. Open the UCI Machine Learning Repository homepage.
  2. Select "Contribute a Dataset."
  3. Complete the submission form — typically this asks for dataset name, description, task type, feature and instance counts, and contact information.
  4. Upload your data files and description document.
  5. Submit and wait for review.

Expected result: after submission, your dataset enters a review process. If accepted, it is published on the repository and becomes publicly accessible, joining the 689 datasets currently maintained.

Common sticking points:

  • The homepage does not list accepted file formats or size limits. Check the contribution page itself for current constraints.
  • There is no published review timeline on the homepage. Do not assume a fast turnaround.
  • If your description is thin, reviewers may need to ask follow-up questions, which lengthens the process. A complete description document reduces back-and-forth.

After Publication

Once published, your dataset appears with a title, task type, instance count, and feature count, following the format used for entries like Iris (Classification, 150 Instances, 4 Features) and Breast Cancer Wisconsin (Diagnostic) (Classification, 569 Instances, 30 Features). Users can then find it through the repository's browsing and search. Because the repository states its datasets are "used by millions of people all around the world," a published dataset can reach a broad audience — which is also why accurate documentation and clear licensing matter before you submit.