Portfolio Projects for Data Analyst Career Changers
Six data analyst portfolio projects for career changers, built on public data from the industry you left: retail, HR, healthcare, teaching, finance, logistics.
The strongest portfolio project you can build as a data analyst career changer sits in the industry you are leaving. A former store manager can download 541,909 real transactions from a UK online gift retailer tonight (UCI Machine Learning Repository). A former recruiter can open monthly federal data showing 7.3 million unfilled jobs in July 2026 (BLS JOLTS). A former teacher can compare program-level debt against first-year earnings at nearby colleges, from files the Department of Education updated in October 2026 (College Scorecard). The pattern is the same each time: public data from your old field, a question your old team would recognize, and interpretation only an insider can produce. That combination is the one clear advantage a career changer holds over every junior analyst who picked a dataset off a leaderboard.
Why your old industry beats a famous datasetPermalink to “Why your old industry beats a famous dataset”
Most beginner advice says to grab Titanic or house prices from Kaggle and practice cleaning. The output is a pile of near-identical notebooks that prove tool familiarity and nothing else. What hiring teams struggle to find in junior candidates is not syntax. It is judgment: knowing which question matters, which number looks wrong, and what the business should do about it on Monday.
Domain knowledge supplies that judgment. You spent years inside retail margins, ward routines, or admissions cycles, and analysts who change careers usually move into roles that build on overlapping skills rather than starting from zero. When your chart shows a spike in cancellations every January, you already know whether that is returns season, wholesale ordering cycles, or a broken feed. A 22-year-old with a fresh certificate has to guess, and interviewers can hear the guessing.
My own switch from sysadmin work into HealthTech analytics turned on exactly this. The project interviewers kept returning to was built on logs from systems I had personally kept running, and it held up because whenever a number looked odd, I could name the operational cause behind it. Nobody else in the applicant pile could do that.
There is an evidence layer to this as well: artifacts from your current job, framed correctly, count as portfolio proof. That argument is covered in past work experience as portfolio evidence. This article covers the build side: new projects, on public data, from the field you already know.
What a domain project needsPermalink to “What a domain project needs”
Before the list, the test. A domain project works when it has three things: a question your former colleagues would actually ask, a dataset from a source you can name and link, and a written verdict a manager could act on. If any of the three is missing, you have an exercise, not a project.
How many projects you need in total, and how they fit into a job search calendar, is strategy territory. The complete portfolio guide for career changers covers that layer. Below, each row pairs a background with the public dataset that background unlocks.
| Your background | Public dataset | The question you already know how to ask |
|---|---|---|
| Retail, store management | UCI Online Retail (541,909 transactions) | Which customer segments drive cancellations, and what would fixing them be worth? |
| HR, recruiting | BLS JOLTS (monthly, by industry) | Which industries kept hiring through the cooldown, and where did quits recover first? |
| Nursing, clinical work | MIMIC intensive care records (credentialed) | What drives length-of-stay differences by admission route? |
| Teaching, education | College Scorecard (updated October 2026) | Which programs at regional colleges leave graduates with debt above first-year earnings? |
| Finance, accounting | FRED (850,000+ series) | How does a rate move flow through to a sector's sales, and with what lag? |
| Logistics, operations | BTS airline on-time database | Which carriers' delays are self-inflicted rather than structural? |
The six projectsPermalink to “The six projects”
1. Retail: the Online Retail datasetPermalink to “1. Retail: the Online Retail dataset”
UCI's Online Retail dataset holds 541,909 transactions from a UK-based online gift retailer, recorded between December 2010 and December 2011, and a large share of the customers are wholesalers. That last fact is where your background starts paying. Wholesale orders distort every average in the file, and invoice numbers beginning with the letter "c" mark cancellations, so returns are recoverable without guessing. Build a customer segmentation that separates wholesale from retail behavior, then quantify what cancellations cost by segment. Close with a one-page memo recommending one change to the returns process, written the way a district manager would read it. SQL or pandas both work; the memo is the part that gets you hired.
2. HR: the JOLTS surveyPermalink to “2. HR: the JOLTS survey”
The Bureau of Labor Statistics runs the Job Openings and Labor Turnover Survey, which publishes monthly national estimates of openings, hires, and separations, plus state-level figures. Job openings stood at 7.3 million in July 2026. You spent years explaining to managers why roles stay open and what quits really cost, so ask the question a recruiting lead would: which industries kept hiring through the 2022 to 2024 cooldown, and where did quits recover first? The deliverable is a turnover briefing written for an HR director, which is the audience you used to be. This is also the rare project where your old job title strengthens the byline instead of needing explanation.
3. Nursing: MIMIC critical care recordsPermalink to “3. Nursing: MIMIC critical care records”
PhysioNet hosts MIMIC, a family of de-identified intensive care databases used in published clinical research. Access is credentialed: you complete a human subjects research course and sign a data use agreement before downloading. Treat that barrier as a feature. The credentialing shows data governance discipline, which matters for anyone hiring an analyst who will touch clinical data, and the questions are ones you can answer from the floor: how length of stay varies by admission route, which patterns in charted vitals precede escalation. Keep the analysis descriptive and conservative. In clinical data, knowing what not to claim is the skill being tested.
4. Teaching: College Scorecard filesPermalink to “4. Teaching: College Scorecard files”
The Department of Education's College Scorecard data publishes field-of-study files with cumulative debt at graduation and earnings one year after graduation for each credential and program combination, last updated in October 2026. You watched students choose colleges on rankings and reputation while the debt and earnings numbers sat in a public file. Pull the five institutions your students actually attended and answer the question guidance counselors get asked every spring: which programs leave graduates owing more than they earn in year one. The deliverable is a single page a counselor could hand to a family. It is a rare artifact that shows both analysis and audience awareness.
5. Finance: FRED time seriesPermalink to “5. Finance: FRED time series”
The St. Louis Fed's FRED database serves over 850,000 economic time series from 127 sources. You know reporting cycles, seasonality, and what a policy rate change does to real activity, so use that. Take one question, such as how retail sales respond to rate moves and with what lag, and answer it properly: align the series to a common frequency, difference what needs differencing, test a couple of lag structures, and write the result as a briefing note. The failure mode you will avoid, because you have read a hundred of these, is presenting correlation in levels between two trending series as if it meant something. That instinct is the portfolio.
6. Logistics: BTS on-time performancePermalink to “6. Logistics: BTS on-time performance”
The Bureau of Transportation Statistics publishes an on-time delay database that splits airline delays into five cause categories, from air carrier to late-arriving aircraft. Operations people spend their careers separating controllable variation from structural variation, and this dataset is that distinction in raw form. Ask which carriers' delays are self-inflicted versus weather or airspace driven, and whether the mix shifted as volumes recovered. SQL plus one honest chart is enough. The write-up should read like the exception report you always wished the tracking system produced.
Turning projects into interviewsPermalink to “Turning projects into interviews”
The project does the heavy lifting, but presentation decides whether it lands. On your resume, each project gets one entry naming what you built, the data source, and one measurable result; the exact format is in how to present side projects on your resume. In interviews, rehearse a 30-second version: the question, the data, the finding, and the recommendation, in that order. Domain projects make this easy, because the recommendation is the part you used to own.
Slot the domain project into a broader structure: the three-project data analytics portfolio covers the other two archetypes, a SQL deep dive and a statistical notebook, and together the three cover the standard interview loop. And before you apply, know what the role pays at each level; the data analyst salary breakdown tracks junior through senior ranges so you negotiate with numbers rather than hope.
Four mistakes that undo domain projectsPermalink to “Four mistakes that undo domain projects”
Defaulting to famous datasets. Titanic clones signal tutorial completion. The datasets above are no harder to obtain and they let you speak with authority in the follow-up questions.
Scrubbing your background. Some changers hide their old career to look more technical. That deletes the exact asset this whole approach is built on. Your title changes; your industry fluency compounds.
Collecting projects instead of finishing them. Six started dashboards lose to one finished analysis with a written verdict. Finish, publish, move to the next.
Shipping charts instead of a decision. Every project above ends in a memo or briefing, not a dashboard link. Reviewers skim conclusions first. Give them one.
How Traecta fits inPermalink to “How Traecta fits in”
Traecta — Your Personalized Career Roadmap starts from your background rather than a generic syllabus. Describe nine years in retail merchandising or a decade of ward nursing, and it names the project from your field that closes your specific gaps against live data analyst requirements, then schedules the SQL and statistics around that build instead of making you redo lessons your old job already taught you.
What to do nextPermalink to “What to do next”
- Pick the row in the table that matches your background and download the dataset this week.
- Write the question and the intended audience before opening the file. If you cannot state both in one sentence each, refine until you can.
- Finish that one project to memo standard, then add the SQL and statistics archetypes around it.
Three focused projects, one of them unshakably yours, will carry a career change further than any volume of generic notebooks. If you want the sequencing done for you, your personalized career roadmap from Traecta maps your work history against data analyst requirements and tells you which project to build first.
SourcesPermalink to “Sources”
- Chen, D. (2015). Online Retail [Dataset]. UCI Machine Learning Repository. archive.ics.uci.edu
- Job Openings and Labor Turnover Survey. U.S. Bureau of Labor Statistics. bls.gov/jlt
- College Scorecard Data. U.S. Department of Education. collegescorecard.ed.gov/data
- Federal Reserve Economic Data (FRED). Federal Reserve Bank of St. Louis. fred.stlouisfed.org
- MIMIC documentation and credentialing. MIT Laboratory for Computational Physiology. mimic.mit.edu
- Airline on-time delay causes. Bureau of Transportation Statistics. transtats.bts.gov
