One of the most common things that stalls people who are learning data analysis is not the tools and not the concepts. It is the data. You finish a tutorial, the sample dataset disappears with the course, and you are left wondering where to find something real to practice on. You want to build a portfolio project but you do not know where to get data that is interesting enough to be worth analyzing and clean enough that you are not spending three weeks just trying to understand what the columns mean.
The good news is that the amount of high quality, freely available data has grown significantly. Governments, research institutions, companies, and communities have published millions of datasets that you can download right now and use in personal, educational, and portfolio projects without paying anything or asking anyone’s permission.
This guide covers the best sources for free datasets organized by what kind of project you are building, with direct links, what each source is best for, and what to watch out for before you start.
What to Look for in a Dataset Before You Download It
Not every free dataset is worth your time. Before committing to a dataset for a portfolio project specifically, check a few things.
Row count matters. A dataset with 200 rows is practice. A dataset with 50,000 rows is a project. Hiring managers and interviewers want to see that you have worked with data at a scale that resembles reality. Aim for at least 10,000 rows for anything you plan to showcase.
Column documentation matters. A dataset where you cannot figure out what half the columns mean from the name alone is a red flag. Look for datasets that come with a data dictionary or a README that explains each field. Time spent reverse-engineering column definitions is time not spent actually analyzing.
Real-world context matters. A dataset that connects to a business problem is more impressive in a portfolio than one that is purely academic. Sales data, customer behavior data, healthcare outcomes, financial transactions, and public infrastructure data all map directly to problems companies are trying to solve.
Format matters for your workflow. Most datasets are available as CSV files, which work with SQL, Python, Excel, R, and virtually every other tool. Some datasets come in JSON or XML format, which requires more preprocessing. If you are just starting out, stick to CSV datasets until you are comfortable with the basics.
Kaggle Datasets
Kaggle is the first place most data analysts go for datasets and for good reason. It hosts hundreds of thousands of user-submitted and curated datasets across every industry and use case imaginable. You can search by topic, filter by file size and format, and see what projects other people have already built on each dataset to get ideas for your own analysis.
The datasets most worth knowing about on Kaggle for portfolio projects include the Sample Superstore sales dataset for retail analysis, the IBM HR Analytics Employee Attrition dataset for people analytics, the NYC Taxi Trips dataset for large-scale transportation analysis, the Black Friday Sales dataset for customer behavior and demographic analysis, and the Titanic dataset for anyone stepping into machine learning for the first time.
Download link: kaggle.com/datasets
What it is best for: SQL projects, Python analysis, machine learning, and portfolio projects across almost every industry.
What to watch out for: Kaggle requires a free account to download datasets. Dataset quality varies significantly because they are user-submitted. Always check the usability score and read the comments before committing to a dataset for a serious project.
UCI Machine Learning Repository
The UCI Machine Learning Repository is one of the oldest and most trusted sources of free datasets for data analysis and machine learning. The datasets here have been used in thousands of published research papers, which means they are well-documented, consistently structured, and thoroughly validated.
The Online Retail dataset from UCI is one of the best datasets available anywhere for e-commerce analysis. It contains over 500,000 rows of transactional data from a UK-based online retailer, with customer IDs, product descriptions, quantities, prices, and dates. It is the foundation for cohort analysis, customer lifetime value analysis, and RFM segmentation projects.
The Adult Census Income dataset, the Wine Quality dataset, and the Heart Disease dataset are all classics for anyone building their first classification model or practicing exploratory data analysis.
Download link: archive.ics.uci.edu/datasets
What it is best for: Machine learning projects, classification analysis, and any project that benefits from a well-documented, research-grade dataset.
What to watch out for: Some datasets are older and may reflect historical rather than current patterns. That is fine for learning purposes but worth noting if you are using the data to make any kind of real-world inference.
Google Dataset Search
Google Dataset Search works exactly like a Google search but specifically for datasets. You type what you are looking for and it returns datasets from across the web including government portals, research institutions, and open data platforms. It covers an enormous range of topics including economics, health, education, environment, and social indicators.
This is the best tool to use when you have a specific domain in mind for a project but do not know where to find the data. If you want to analyze air quality data, unemployment trends, school performance, or hospital capacity, Google Dataset Search will surface sources you would not find by browsing Kaggle alone.
Download link: datasetsearch.research.google.com
What it is best for: Domain-specific research, finding niche datasets on topics not well covered by general repositories, and government or institutional data.
What to watch out for: Results vary enormously in quality and format. Some datasets require registration or have usage restrictions. Always check the license before using a dataset in a public project.
Data.gov
Data.gov is the United States government’s open data platform and it contains over 290,000 datasets from federal agencies. This is where you find data on public health, education, infrastructure, environment, economics, agriculture, and dozens of other domains. Every dataset here is produced and maintained by a government body, which means the data collection methodology is documented and the sourcing is credible.
Useful datasets on Data.gov for portfolio projects include the Food Environment Atlas for analyzing the relationship between food access and health outcomes, the Federal Aviation Administration flight delay data for transportation analysis, the NYPD complaint data for public safety analysis, and the Bureau of Labor Statistics employment data for economic analysis.
Download link: data.gov
What it is best for: Projects with a civic, economic, or public health angle. Data that needs to be credibly sourced for any kind of published analysis.
What to watch out for: Government datasets can be messy. Missing values, inconsistent formatting across years, and cryptic agency codes are common. This actually makes them excellent for practicing data cleaning but factor in extra preparation time before analysis.
Maven Analytics Data Playground
Maven Analytics maintains a curated collection of free datasets specifically designed for data analysis practice and portfolio projects. What makes this source different from the others is that the datasets are hand-picked by instructors who know exactly what skills they are designed to demonstrate. Each dataset comes with a clear description and suggested project ideas.
Notable datasets available on Maven Analytics include NYC taxi trip records for large-scale transportation analysis, Kickstarter project data with over 375,000 campaigns for success prediction and trend analysis, historical Olympic Games data from 1896 to 2016 for sports analytics, customer churn data for a fictional telecom company for retention analysis, and S&P 500 historical stock data for financial analysis.
Download link: mavenanalytics.io/data-playground
What it is best for: Beginners who want clean, well-documented datasets with clear project direction. Anyone building their first portfolio project and wanting to spend time on analysis rather than data cleaning.
What to watch out for: The collection is smaller than Kaggle so the selection is more limited. For niche topics you may need to go elsewhere.
Our World in Data
Our World in Data is a research publication from the University of Oxford that publishes data-driven articles on global issues including health, poverty, education, energy, and climate change. Every chart and statistic on the site is backed by a downloadable dataset.
What makes this source particularly valuable for data analysts is the combination of data quality and real-world significance. The COVID-19 dataset published by Our World in Data became one of the most analyzed public datasets of the last decade. The energy mix dataset, the global life expectancy data, and the CO2 emissions dataset are all excellent foundations for projects that demonstrate both analytical skill and awareness of meaningful global issues.
Download link: ourworldindata.org/data
What it is best for: Projects with a global or social impact angle. Time-series analysis across countries. Any project where the context and credibility of the data source matters as much as the analysis itself.
What to watch out for: Datasets are country-level and macro in nature. They are not suitable for customer-level or transaction-level analysis. Use them for trend analysis, comparisons, and time-series projects.
GitHub Open Datasets
GitHub hosts a large number of open datasets directly in repositories, and the Awesome Public Datasets repository maintained at github.com/awesomedata/awesome-public-datasets is the most comprehensive categorized list of free public datasets available anywhere online. It covers agriculture, biology, climate, economics, education, finance, healthcare, sports, transportation, and dozens of other categories, each with direct links to the source data.
Some of the most useful datasets hosted directly on GitHub include data on every member of the US Congress from 1789 onwards, Chicago food inspection records, New York City open data exports, and various sports statistics datasets across football, basketball, and baseball.
Download link: github.com/awesomedata/awesome-public-datasets
What it is best for: Finding datasets in specific domains that general repositories do not cover well. Discovering niche datasets that make for more original portfolio projects.
What to watch out for: Dataset quality and documentation varies widely. Some repositories have not been updated in years. Check the last commit date and the README before investing time in a dataset.
Five38 and BuzzFeed News
FiveThirtyEight publishes datasets from the political and sports analysis that appears in their articles, all freely available on their GitHub repository. If you want to practice political data analysis, sports statistics, or economic polling data, this is one of the cleanest and most interesting collections available.
BuzzFeed News similarly published datasets from their investigative journalism work, including data on flight tracking, government spending, and social media analysis. These datasets are notable because they were used in real published journalism, which gives them a level of credibility and real-world context that practice datasets often lack.
Download link: github.com/fivethirtyeight/data and github.com/BuzzFeedNews
What it is best for: Political and sports analysis projects. Journalism-oriented data work. Projects where an interesting or counterintuitive finding is the goal.
What to watch out for: Some datasets are tied to specific articles or events from several years ago. The analysis context from the original article can be helpful for understanding what questions the data was originally collected to answer.
Free Datasets Cheat Sheet
| Source | Best For | Format | Row Count |
|---|---|---|---|
| Kaggle | SQL, Python, ML, all industries | CSV | Varies, 1K to millions |
| UCI Repository | Machine learning, classification | CSV | 1K to 500K+ |
| Google Dataset Search | Domain-specific and niche topics | CSV, JSON | Varies |
| Data.gov | Civic, health, economic analysis | CSV, Excel | Thousands to millions |
| Maven Analytics | Beginner-friendly, clean datasets | CSV | 10K to 500K |
| Our World in Data | Global trends, time-series | CSV | Country-level |
| GitHub Awesome Datasets | Niche and specialized domains | CSV, JSON | Varies |
| FiveThirtyEight | Politics, sports, economics | CSV | 1K to 100K |
Common Mistakes to Avoid
Choosing a dataset before choosing a question. A dataset without a question is just a file. Start with the business question you want to answer, then find the dataset that contains the variables you need to answer it. Going the other way almost always leads to unfocused analysis that does not land well in a portfolio.
Using the most popular datasets without differentiation. The Titanic dataset and the Iris dataset have been analyzed millions of times. Using them without adding something original means your project looks identical to thousands of others. If you use a popular dataset, ask a question the standard tutorials do not ask.
Skipping data exploration before building the project. Before writing a single analytical query or line of code, spend time understanding the dataset. How many rows? What are the date ranges? Are there NULL values? Are there columns that look numeric but contain text? That exploration shapes every analytical decision you make afterward and catching data quality issues early saves significant rework time.
Downloading data and never doing anything with it. The common behavior among people learning data analysis is accumulating datasets like hobbies instead of actually building projects with them. Pick one dataset, define a specific question, and finish the analysis before downloading another. A completed project on one dataset is worth more than ten downloaded datasets that were never analyzed.
Finding good data is genuinely half the battle in data analysis work. The other half is knowing what questions to ask once you have it. The sources on this list give you access to the same quality of data that professional analysts work with every day, with no budget and no permissions required. The only thing between you and the project is starting.
FAQs
Where can I find free datasets for data analysis?
The best sources for free datasets are Kaggle, the UCI Machine Learning Repository, Data.gov, Maven Analytics Data Playground, Codewithfimi, Our World in Data, and Google Dataset Search. Each source has different strengths. Kaggle is the broadest and most community-driven. UCI has the most research-validated datasets. Data.gov has the most credible government data. Maven Analytics has the cleanest beginner-friendly options.
What format are most free datasets available in?
The majority of free datasets are available as CSV files, which work with SQL, Python pandas, R, Excel, and virtually every other data analysis tool. Some datasets come in JSON or XML format, which requires additional preprocessing. When choosing a dataset for a beginner project, filter for CSV format to minimize the setup work before you can start the actual analysis.
How big should a dataset be for a portfolio project?
For a portfolio project, aim for a dataset with at least 10,000 rows. This is large enough to produce meaningful aggregations, demonstrate that you can work with real-world data volumes, and show patterns that are statistically meaningful. Datasets with 100,000 to 500,000 rows are ideal for intermediate projects that include window functions, cohort analysis, or machine learning.
Can I use free public datasets for commercial projects?
It depends on the license. Most datasets from government sources like Data.gov are in the public domain and can be used freely for any purpose. Kaggle datasets vary by license and each one specifies its terms on the dataset page. UCI datasets are generally available for educational and research use. Always check the license before using any dataset in a published or commercial project.
What is the easiest dataset to start with for a first project?
The Sample Superstore dataset on Kaggle is the most beginner-friendly starting point for data analysis. It is clean, well-documented, has intuitive business columns including sales, profit, category, and region, and is large enough at over 9,000 rows to produce meaningful analysis. It works well with SQL, Python, Excel, and any BI tool, making it flexible for whatever tool you are learning.