Final-year data project ideas that aren't clichés, with a rubric
In this article (5 sections)
Related programme: the Data Science course (6 months; ₹1,30,000 (inclusive of GST) · EMI from ₹5,000/month).
A final-year data project stands out when it answers a specific decision for a named user, compares against a simple baseline, and states what it cannot do — not when it uses the most advanced model. The eight ideas below are built that way, and the 100-point rubric at the end lets you score your own project before the viva and before an interviewer does it for you.
Why familiar projects fall flat
Projects such as Titanic survival, iris classification, house prices and movie recommenders are fine for learning. The problem is that they arrive with a solved dataset, a known answer and no decision attached, so there is nothing for an interviewer to ask beyond "which model did you use?". You can often rescue one with a change of question:
| Cliché | Upgrade with a decision |
|---|---|
| Titanic survival | Lifeboat allocation: given limited seats, which rule saves the most lives, and how does it trade off fairness? |
| House price prediction | Should a buyer negotiate? Flag listings priced above the model's range, and measure how often the flag is wrong. |
| Movie recommender | For a college library: which ten books to buy with a fixed budget, using issue records |
| Sentiment analysis of tweets | Triage your college's feedback form: which complaints need a same-week response? |
| Stock price prediction | Drop it. A price forecast without a trading-cost backtest does not survive questioning. |
Eight ideas built around a decision
Each idea names the user, the decision and the baseline you must beat. Data sources are suggestions; confirm availability and licences before you commit.
- 1Hostel mess waste forecasting. User: mess manager. Decision: how many meals to prepare per day. Baseline: last week's same-day count. Data: collect daily headcounts and waste weights for a few weeks, with the warden's permission.
- 2Library seat-occupancy prediction. User: librarian. Decision: when to open an overflow room during exams. Baseline: fixed exam-week schedule. Data: hourly entry counts, logged manually or from a gate register.
- 3Bus-delay flagging for a college route. User: transport office. Decision: which departures to reschedule. Baseline: current timetable. Data: logged arrival times over a month.
- 4Placement-cell document checker. User: placement coordinator. Decision: which student applications are incomplete before submission. Baseline: manual checklist. Data: synthetic resumes and forms; never real student records without consent.
- 5Crop price alert for a local market. User: a farmer deciding when to sell. Decision: sell now or wait a week. Baseline: sell immediately. Data: check whether your state's agricultural marketing board or the government's open data portal (data.gov.in) publishes mandi prices for your crop.
- 6Electricity bill anomaly detector. User: a housing society. Decision: which flats' bills to verify. Baseline: flag anything 50% above last month. Data: anonymised bills with the society's agreement, or synthetic.
- 7Question-bank difficulty estimator. User: a teacher. Decision: which questions to revise before reuse. Baseline: teacher's own difficulty label. Data: anonymised past test responses with the department's approval.
- 8Grounded FAQ assistant for your department. User: first-year students. Decision: answer from the handbook or say "ask the office". Baseline: keyword search. Evaluation: 50 questions with correct answers written by you, scored for correctness and correct refusals.
Notice what these share: a person who would use the output, a cheaper alternative you must beat, and a way to measure being wrong.
The 100-point rubric
Score your project honestly, or better, ask a classmate to score it.
| Area | Points | What full marks looks like |
|---|---|---|
| Problem and user | 15 | One named user, one decision, and why it matters to them, in under 100 words |
| Data | 15 | Source documented, collection method explained, consent or licence stated, known gaps listed |
| Baseline | 15 | A simple rule or naive forecast implemented and scored on the same test data |
| Method | 15 | Choice justified against the baseline; no step that leaks test data into training |
| Evaluation | 15 | Metric tied to the cost of mistakes; results shown on held-out data; errors examined by type |
| Limits and ethics | 10 | What the project cannot do, who could be harmed by a wrong output, and how you checked |
| Reproducibility | 10 | Another student can rerun it from the README in under 30 minutes |
| Communication | 5 | A one-page summary a non-technical user could act on |
Reading your score: under 50, the project needs a baseline and a clearer user before anything else. From 50 to 74, it will pass a viva but may struggle in a technical interview; focus on evaluation and limits. At 75 and above, you have a project you can put first on your resume.
A worked scoring example
A hypothetical student builds the mess-waste forecast with a gradient-boosting model. Scoring: problem and user 14 (clear); data 10 (collected for 25 days, but two missing weeks not explained); baseline 5 (mentioned but not computed); method 12; evaluation 8 (only overall error, no look at festival days); limits 4; reproducibility 9; communication 4. Total: 66.
The fastest improvements: compute the last-week baseline (+10 possible), analyse the days with the largest errors (+5), and write the limits section, including that 25 days cannot capture exam or festival patterns (+5). That is three short tasks taking the project past 80 without changing the model.
Before your viva
Prepare answers to four questions examiners and interviewers return to: why this metric, how much better than the baseline and is it worth the complexity, what would make the model wrong, and what you would do with another month. If you already have a smaller college project, making a minor project interview-ready applies the same rubric in a shorter form. For the timing of the project within your degree, see the semester-by-semester plan.
Baselines, leakage checks and error analysis are the parts of a project that take longest to learn alone; they are assessed on every project in the Data Science programme.
Frequently asked questions
Is Titanic or house price prediction acceptable as a final-year project?
Your college may accept it, but interviewers have seen it many times and it rarely leads to a good conversation. If you must use a familiar dataset, change the question so there is a decision and a cost of being wrong, and show a baseline comparison.
Can I use synthetic data for a final-year project?
Yes, if you say so clearly and explain how you generated it. Synthetic data is sensible when real data contains personal information. Pair it with at least one real public dataset if your guide expects real-world evidence.
Pankit Kumar has 10 years in Data Science & AI, building and shipping production systems in regulated pharma and clinical environments. He is a freelance trainer at Boston Institute of Analytics, AnalytixLabs and Scaler, and has taught this material to thousands of working professionals.
This article is part of our Data Science programme — 6 months. From data foundations to machine learning, deep learning and deployment.
Explore Data Science