GhostpilotHome · Interview questions · Data Scientist
Data Scientist interview questions
Data science interviews cover statistics, modelling and product sense. You'll be judged on how you frame a problem and test whether your result is real, as much as on the model you pick.
What interviewers look for
- Statistics. Experiments, significance and bias.
- Modelling judgement. Choosing simple baselines and the right metric.
- Business impact. Connecting a model to a decision.
How the interview usually runs
- Recruiter call.
- SQL and Python screen.
- Statistics and machine learning interview.
- Case study, then a behavioural round.
Stages vary by company, so ask the recruiter what to expect before you start preparing.
10 Data Scientist interview questions, and how to answer them
1. Explain a p-value to a non-technical person.
What they want to see: Clear communication of statistics.
How to answer: It's how surprising the data would be if nothing had changed. It isn't the chance your idea is right.
2. Precision or recall?
What they want to see: Choosing metrics.
How to answer: Decide from the cost of false positives against false negatives, with an example.
3. How do you detect overfitting?
What they want to see: Validation.
How to answer: Compare training and validation scores, and use cross-validation and regularisation.
4. How do you handle class imbalance?
What they want to see: Practical modelling.
How to answer: Use the right metric, then resampling, class weights or threshold tuning.
5. How would you design an A/B test?
What they want to see: Experiment design.
How to answer: Define the metric, power and sample size, randomise, and don't stop early.
6. Random forest or gradient boosting?
What they want to see: Model knowledge.
How to answer: Compare bagging and boosting, tuning effort and overfitting risk.
7. What is data leakage?
What they want to see: Rigour.
How to answer: Information from the future or target reaches training. Split by time and check suspiciously good results.
8. How do you explain a model's decision?
What they want to see: Trust.
How to answer: Use feature importance and SHAP, and show clear examples.
9. How would you build a churn model?
What they want to see: End-to-end thinking.
How to answer: Define churn, build features, set a baseline and connect the output to a retention action.
10. Tell me about a model that failed.
What they want to see: Honesty.
How to answer: Say what went wrong and what you changed.
A sample answer
How would you build a churn model?
I'd start by agreeing what churn means and what the team can do about it. At a subscription company, I used usage, tenure and support data, with logistic regression as a baseline and gradient boosting after. We targeted the top decile with a retention offer in a controlled test. That group churned noticeably less, which justified keeping the model.
A made-up example to show the structure. Use your own experience and real numbers.Common mistakes
- Reaching for deep learning before a baseline.
- Quoting accuracy on imbalanced data.
- Unable to explain statistics plainly.
- No link between the model and the business.
Practise this out loud
Run a mock interview for a Data Scientist role. Questions change every time, you answer by voice or typing, and each answer gets a score with feedback.
Frequently asked questions
How much maths do data scientists need?
Be comfortable with probability, statistics and basic linear algebra. Applied roles care more about using them well than proofs.
Is SQL tested for data scientists?
Yes, usually. Many roles spend a large share of time on data access, so expect a SQL screen.