15 Data Science Assignment Topics for University Students

Your topic shapes your grades, whether it includes house prices, NHS waiting lists, or fraud alerts. A data science assignment topic is a focused question that real data, a clear method and a measurable result can answer. Weak data or a vague aim holds back even strong code. Here this guide provides you 15 data science assignment topics, each built on a real, openly accessible dataset with a method comparison and the critical angle UK markers reward at 2:1 and first-class level. Check licensing and access for any dataset before your commit.

What Separates a 2:1 Topic From a First-Class One

Your topic sets the ceiling on your marks, so judge it first. Usually UK universities place a first class at 70% and above and a 2:1 between 60% and 69%. At the end, markers reward critical judgement on the basis of justified methods, a fair baseline, honest evaluation and clear limitations. A built-in comparison lets you show all of that.

Check the data next. As much UK public data uses the Open Government Licence, but confirm the terms. If the data covers people, UK GDPR applies, and many universities want even secondary data declared for ethics, so ask your tutor. Then only match the scope to your word limit. Because one question and two models done properly beat five done lightly.

Each topic below names a dataset, a comparison and a critical angle, so you can start building straight away.

Machine Learning and Prediction Topics

These below-listed suit modules are focused on modelling and evaluation.

• House Price Prediction with HM Land Registry Data: Compare linear regression and gradient boosting with a time-based split, since prices drift. Explaining why a random split flatters results earns first-class credit.
• Fraud Detection on Imbalanced Transactions: Start by using the ULB credit card dataset on Kaggle. Compare class weights with SMOTE, judge by PR-AUC, and justify your threshold using the cost of a missed fraud.
• Customer Churn Prediction: Pick the IBM Telco churn dataset. Compare logistic regression with XGBoost, then explain predictions with SHAP values and link top drivers to a retention plan.
• Student Dropout Prediction: Go with the UCI dropout and academic success dataset. Check whether error rates differ across groups such as gender, and discuss the ethics of acting on predictions.
• Heart Disease Risk Prediction: Use the UCI Heart Disease data. Prioritise recall, since missed cases cost more than false alarms, and state the limits of a small, older sample.

These test your modelling skills, while the next group tests how you handle official UK data.

UK Public Data and Time Series Topics

In these topics official sources add credibility, but they also bring messy formats.

• NHS A&E Attendance Forecasting: Use NHS England's monthly A&E statistics. Compare SARIMA with Prophet and test how each handles the sharp COVID-19 drop in 2020.
• Road Accident Severity Prediction: Use Department for Transport STATS19 data. Handle the rarity of fatal cases, then discuss reporting bias, since the data covers only police-reported accidents.
• Air Quality Forecasting for UK Cities: Use nitrogen dioxide readings from DEFRA UK-AIR. Beat a persistence baseline, where tomorrow equals today, and explain how you handled missing readings.
• Crime Hotspot Analysis: Use street-level data from data.police.uk. Apply DBSCAN, adjust for population so city centres do not dominate, and note that locations are anonymised.
• Inflation Drivers in the UK: Use ONS consumer price index data by category. Run stationarity tests before regression to avoid spurious links, then interpret the role of energy and food.

Once you can clean official data, you are ready for text, ethics and deeper methods.

Text, Ethics and Advanced Method Topics

These suit final-year work and modules that reward critical discussion.

• Sentiment Analysis of Customer Reviews: Use the IMDb review dataset. Compare VADER with a fine-tuned DistilBERT model and run an error analysis on sarcasm and negation.
• Topic Modelling of Hansard Debates: Use Hansard transcripts from UK Parliament. Compare LDA with BERTopic using coherence scores, then validate sample topics by reading the actual speeches.
• Fairness in Loan Approval Models: Use the UCI German Credit dataset. Measure demographic parity and equalised odds, and relate findings to the Equality Act 2010 and UK GDPR Article 22.
• Image Classification with Transfer Learning: Use CIFAR-10. Compare a small CNN built from scratch with a fine-tuned ResNet, and use learning curves to show where each overfits.
• Movie Recommender System: Use the MovieLens dataset. Compare collaborative and content-based filtering using NDCG, and explain how your system treats new users with no history.

With a shortlist ready, test it against the checks below.

Before You Lock In Your Topic

Use this list to test whether your shortlist is realistic.

• Can you download and load the dataset today without errors.
• Does the licence allow academic use, and have you noted it for your references.
• Can you build a simple baseline model within the first week.
• Does the topic fit your word limit with room for discussion and limitations.
• Has your tutor approved it, especially if personal data is involved.

If your pick passes, you are ready to commit.

Final Word

The strongest topic is the one where you can show a result early. Load the data and draw one chart this week. If the data fights you now, switch before you lose three weeks. You can also prefer taking data science assignment support for saving time and efficient project preparation. Also, a smaller question with a clear, well-argued answer will usually score higher than an ambitious one you cannot finish.