Skip to content
JOBSPOT 360 Career guidance, interview preparation and genuine jobs

Data Scientist / ML Engineer interview questions

Data science and ML interviews combine statistics, machine learning theory, coding in Python and an applied case. Interviewers look for people who understand why a model works, how to evaluate it honestly and how it would behave in production.

11 questions with answer notes · 4 rounds · By the JOBSPOT 360 team

How the selection process usually works

  1. 1

    Online assessment

    Python coding, SQL, probability and ML MCQs.

  2. 2

    ML fundamentals

    Algorithms, bias-variance, metrics, feature engineering and model evaluation.

  3. 3

    Case study or take-home

    Build or design a model for a business problem and present your approach.

  4. 4

    Project deep-dive and HR

    Your projects or research in depth, deployment, then HR.

Rounds vary between companies. Smaller teams often merge the technical and managerial rounds, while larger companies may add an extra technical round.

Skills to brush up on

  • Python (NumPy, pandas, scikit-learn)
  • Statistics and probability
  • Supervised and unsupervised ML
  • Deep learning (PyTorch or TensorFlow)
  • SQL
  • Model evaluation and validation
  • LLMs, embeddings and RAG basics
  • Model deployment (APIs, MLOps basics)

Technical questions

  1. 01 Explain the bias-variance trade-off.

    How to answer: High bias underfits; high variance overfits. Regularisation, more data, a simpler model or ensembles move you along the trade-off.

  2. 02 How do you handle an imbalanced dataset?

    How to answer: Use suitable metrics (precision, recall, F1, PR-AUC), class weights, resampling such as SMOTE, and threshold tuning. Accuracy is misleading here.

  3. 03 Precision or recall: which matters more for fraud detection, and for spam filtering?

    How to answer: Fraud usually favours recall (do not miss fraud). Spam filtering favours precision (do not hide real emails). Tie the choice to the business cost of each error.

  4. 04 How does a random forest differ from gradient boosting?

    How to answer: A random forest averages independent trees (bagging) to reduce variance. Boosting builds trees sequentially, each correcting the last, which reduces bias.

  5. 05 What is data leakage? Give an example.

    How to answer: Training on information that will not be available at prediction time, such as scaling before the train/test split or using a feature recorded after the outcome.

  6. 06 What do L1 and L2 regularisation do?

    How to answer: Both penalise large weights. L1 can push weights to zero (feature selection); L2 shrinks them smoothly.

  7. 07 Explain how a transformer uses attention.

    How to answer: Each token builds queries, keys and values. Attention weights come from query-key similarity and mix the values, letting each token use context from the whole sequence.

  8. 08 What is retrieval-augmented generation (RAG) and when would you use it?

    How to answer: Retrieve relevant documents using embeddings and pass them to an LLM as context. Use it to ground answers in private or fresh data without fine-tuning.

Scenario questions

  1. 01 Design a model to predict customer churn for a telecom company.

    How to answer: Define churn and the prediction window, engineer usage and billing features, pick a baseline then stronger models, choose metrics tied to retention cost, and plan monitoring.

  2. 02 Your model scores 95% on validation but performs poorly in production. Why?

    How to answer: Look for data leakage, distribution drift, training and serving differences in preprocessing, or a validation set that does not match real traffic.

  3. 03 How would you explain your model's predictions to a business team?

    How to answer: Use feature importance, SHAP values and concrete example cases, and focus on what action each prediction enables.
Then comes the HR round Almost every interview ends with HR questions about you, your plans and your salary expectations. See the common ones and how to answer them.

Resume tips for this role

  • For each project, state the problem, the data, the model, the metric and the outcome.
  • Link GitHub notebooks or a deployed demo (Streamlit, Hugging Face Spaces).
  • Kaggle medals, publications or open-source contributions are strong signals.
  • Do not list every algorithm. List what you have applied to real data.

This guide is written by the JOBSPOT 360 team from common interview patterns for this kind of role. It is not provided by any employer, and real interviews will differ. Read our editorial policy.