Free Data Scientist Interview Prep

Data Scientist Interview Questions.

16 practice questions across 11 skills — each with what the interviewer is really listening for. Prepare with intent, not guesswork.

Free · No signup · Grouped by skill

How to use this page

Don't memorise answers. For each question, read what the interviewer is looking for, then practise answering out loud in your own words with a real example. The notes describe strong answers and common red flags — they are for your prep, not a script.

Python

1 question
  1. How would you use Python to explore a new dataset before any modelling?

    Basic

    What the interviewer is looking for: Strong answers describe loading with pandas, checking shape, dtypes, missingness, distributions, and correlations before touching a model. Jumping straight to a model without EDA is a red flag.

Machine Learning

3 questions
  1. Explain the bias-variance trade-off and how it shows up in practice.

    Intermediate

    What the interviewer is looking for: Look for underfitting (high bias) versus overfitting (high variance), and levers like model complexity and regularisation. Reciting the phrase without connecting it to overfitting is a red flag.

  2. How do you decide whether a classification model is actually good, beyond accuracy?

    Advanced

    What the interviewer is looking for: Strong candidates raise precision, recall, F1, ROC-AUC, and the danger of accuracy on imbalanced data. Relying on accuracy alone for a rare-event problem is a serious red flag.

  3. Your model performs great in training but poorly in production. What might be happening?

    Advanced

    What the interviewer is looking for: Look for overfitting, data leakage, train/serve skew, and distribution drift, with a plan to diagnose each. Only suggesting 'more data' without diagnosing leakage or drift is a red flag.

Statistical Analysis

2 questions
  1. What is a p-value, and what does it not tell you?

    Intermediate

    What the interviewer is looking for: A strong answer defines it as the probability of data at least this extreme under the null, and warns it is not the probability the hypothesis is true nor effect size. Misinterpreting it as 'probability the result is real' is a red flag.

  2. How would you check whether two groups differ in a metric in a statistically sound way?

    Intermediate

    What the interviewer is looking for: Look for choosing an appropriate test, checking assumptions, considering sample size and effect size, and multiple-comparison caution. Eyeballing means without any test is a red flag.

SQL

1 question
  1. Write a query to compute a 7-day rolling average of daily active users. Explain your approach.

    Intermediate

    What the interviewer is looking for: Strong candidates reach for a window function with a frame, handle gaps in dates, and reason about performance. Not knowing window functions or ignoring missing dates is a red flag.

Pandas

1 question
  1. How do you use pandas to handle a dataset that is too large to fit comfortably in memory?

    Advanced

    What the interviewer is looking for: Look for chunking, dtype downcasting, selecting only needed columns, or moving to a database/Dask. Insisting on loading everything into one DataFrame is a red flag.

NumPy

1 question
  1. When would you use NumPy operations instead of pandas or plain Python loops?

    Intermediate

    What the interviewer is looking for: A good answer highlights vectorised operations for speed and memory efficiency on numerical arrays, and avoiding Python-level loops. Not understanding vectorization is a red flag.

Matplotlib

1 question
  1. How do you choose the right chart with Matplotlib to communicate a model's results?

    Basic

    What the interviewer is looking for: Look for matching the visual to the message — a confusion matrix, ROC curve, or residual plot as appropriate — and labelling clearly. Decorative charts that do not aid understanding are a red flag.

Feature Engineering

1 question
  1. What is feature engineering, and can you give an example that meaningfully improved a model?

    Advanced

    What the interviewer is looking for: Strong answers show domain-driven features, encoding, and transformations, with a concrete example and its measured effect. Vague claims with no example or leaking the target into a feature are red flags.

A/B Testing

1 question
  1. How would you design an experiment to measure the impact of a new recommendation model?

    Advanced

    What the interviewer is looking for: Look for a clear hypothesis and primary metric, randomisation, power analysis, guardrail metrics, and waiting for significance. Declaring a winner early or lacking a control are red flags.

Scikit-learn

1 question
  1. When would you reach for scikit-learn versus a deep learning framework like TensorFlow or PyTorch?

    Intermediate

    What the interviewer is looking for: Strong candidates favour classical models for tabular and smaller data and deep learning for unstructured data at scale, prioritising the simplest model that works. Reaching for deep learning by default is a red flag.

Data Storytelling

1 question
  1. How do you translate a model's output into a recommendation a business leader can act on?

    Intermediate

    What the interviewer is looking for: Look for framing in business terms, quantifying expected impact, and being honest about uncertainty and assumptions. Presenting model metrics with no business translation is a red flag.

Behavioural & role-general

2 questions
  1. A stakeholder asks for a model, but you suspect the problem does not need one. How do you respond?

    Intermediate

    What the interviewer is looking for: Strong answers probe the underlying business question and are willing to propose a simpler analytical solution. Building a complex model just because it was requested is a red flag.

  2. Tell me about a data science project where the result surprised you. What did you do next?

    Basic

    What the interviewer is looking for: Look for scientific rigour — validating the surprise, checking for errors and leakage, and communicating honestly — rather than either dismissing or over-trusting it. Accepting a surprising result without verification is a red flag.

Before the interview, close the skill gaps first

Upload your resume and see your Data Scientist Skill Score, the exact skills you're missing, and a free learning path for each one — so you walk in prepared.