Data Science Projects for Freshers: 12 Ideas with Datasets

A portfolio project earns attention when it answers a question someone would care about, on data that took real work to use, with a clear write-up of what was found. The twelve ideas below each pair a real, public dataset with such a question. They are grouped by the skill they prove, so you can pick two or three that together show range: one on structured prediction, one on unsupervised or text data, and one that shows you can handle scale or tell a story.

For each, the dataset is linked, and each idea ends with a stretch goal that separates a finished project from a tutorial exercise.

Prediction on business data

1. Which customers will respond to a campaign? Use the Bank Marketing dataset of a bank's telephone campaigns (UCI: Bank Marketing). Build a classifier for subscription, deal with the imbalance between yes and no, and choose a decision threshold from a stated cost of calling versus the value of a subscription. Proves: classification, imbalanced evaluation, business framing. Stretch: show how many calls the model saves for the same number of subscriptions.

2. Which customer segments does a shop have? Use the Online Retail transactions (UCI: Online Retail). Clean returns and missing customer IDs, compute recency, frequency and monetary value per customer, and cluster them. Proves: messy-data cleaning, feature building, clustering. Stretch: name each segment in plain words and propose one action per segment.

3. How many bikes will be rented tomorrow? Use the Bike Sharing data with weather and calendar fields (UCI: Bike Sharing). Model hourly demand, respecting time order in validation. Proves: regression, time-aware splitting, feature engineering. Stretch: compare against a simple same-hour-last-week baseline and report how much the model improves on it.

4. What makes a wine score well? Use the Wine Quality measurements (UCI: Wine Quality). Predict the quality score and interpret which properties matter. Proves: regression versus ordinal classification choices, model interpretation. Stretch: check whether red and white wines need separate models.

Responsible modelling

5. Is an income model fair across groups? Use the Adult census dataset (UCI: Adult). Train a classifier, then compare error rates across groups such as sex and race. Proves: classification plus fairness evaluation, a topic interviewers increasingly raise. Stretch: show how changing the threshold per group changes the trade-offs, and discuss whether that is appropriate.

6. Can a small medical dataset support a trustworthy model? Use the Heart Disease data (UCI: Heart Disease). With a few hundred rows, focus on careful cross-validation, simple interpretable models and honest uncertainty. Proves: validation discipline on small data. Stretch: a one-page limitations section explaining why the model is not fit for clinical use.

Text

7. A spam filter from scratch. Use the SMS Spam Collection (UCI: SMS Spam Collection). Build a TF-IDF and logistic regression classifier and examine its mistakes. Proves: text preprocessing, baseline NLP. Stretch: analyse which words drive spam predictions and where the model fails.

8. Sentiment of movie reviews. Use the IMDb reviews dataset (Hugging Face: IMDb). Compare a bag-of-words baseline with a pre-trained transformer model. Proves: NLP, comparing a baseline with a modern model. Stretch: measure the accuracy gained against the extra compute needed.

Scale and geography

9. When and where do taxi riders tip most? Use New York's trip records (TLC Trip Record Data). Load several months of Parquet files and analyse tips by hour, zone and payment type. Proves: working with tens of millions of rows efficiently. Stretch: repeat one analysis in pandas and in DuckDB or Spark, and compare speed and code.

10. What drives house values, and where does the model fail? Use California housing from scikit-learn (scikit-learn: fetch_california_housing). Model median value and map the residuals. Proves: regression, spatial thinking, error analysis. Stretch: add a feature such as distance to the coast and measure its effect.

Analysis and storytelling

11. How have development indicators changed? Use World Bank Open Data (World Bank Open Data). Pick a question, such as how access to electricity relates to school enrolment over time, and answer it with a few well-made charts. Proves: exploratory analysis, joining indicators, communication. Stretch: a two-page report that a non-technical reader can follow.

12. A question about India, answered with public data. Use the Open Government Data Platform India (data.gov.in). Choose a dataset on a topic you know, clean it, and answer one specific question with evidence. Proves: real-world cleaning and local relevance. Stretch: combine two datasets from different departments to answer something neither answers alone.

Make any of these portfolio-worthy

• Start with the question in the first line of the README, and end with the answer.
• Show the baseline. A model is only impressive relative to a simple alternative.
• Write down the limits: data problems, what the model cannot do, what you would try next.
• Keep it reproducible: a requirements file, a clear folder structure and a notebook that runs top to bottom.

For more places to find data, see free datasets for data science practice; for turning projects into interviews, see how to build a data science portfolio. Several of these ideas echo the capstone projects in the Data Science course, such as churn prediction, sales forecasting and sentiment analysis.

Sources

• Bank Marketing (UCI) — UCI Machine Learning Repository
• Online Retail (UCI) — UCI Machine Learning Repository
• Bike Sharing (UCI) — UCI Machine Learning Repository
• Wine Quality (UCI) — UCI Machine Learning Repository
• Adult (UCI) — UCI Machine Learning Repository
• Heart Disease (UCI) — UCI Machine Learning Repository
• SMS Spam Collection (UCI) — UCI Machine Learning Repository
• IMDb movie reviews — Hugging Face
• TLC Trip Record Data — NYC Taxi and Limousine Commission
• fetch_california_housing — scikit-learn documentation
• World Bank Open Data — World Bank
• Open Government Data Platform India — Government of India