Machine learning interviews usually mix fundamentals, practical model-building questions and, for more senior roles, ML system design. Interviewers want clear explanations and sound judgement about trade-offs, not memorised definitions. Here are common questions with example answers.
Fundamentals
What is the difference between supervised and unsupervised learning?
Supervised learning trains on labelled examples to predict an output, such as classifying emails as spam or not spam. Unsupervised learning finds structure in unlabelled data, such as clustering customers into segments.
Explain the bias–variance trade-off.
Bias is error from overly simple assumptions, which leads to underfitting. Variance is error from being too sensitive to the training data, which leads to overfitting. Making a model more complex usually lowers bias but raises variance. The goal is the balance that gives the lowest error on unseen data.
What is overfitting and how do you prevent it?
Overfitting is when a model performs well on training data but poorly on new data. Common fixes include more training data, simpler models, regularisation (such as L1 or L2), dropout in neural networks, early stopping, and using cross-validation to choose hyperparameters.
Model Evaluation
When would you use precision vs recall?
Precision is the share of predicted positives that are correct. Recall is the share of actual positives you found. Prioritise recall when missing a positive is costly, as in disease screening. Prioritise precision when false alarms are costly, as in flagging transactions for manual review. F1 balances the two.
Why can accuracy be misleading?
On imbalanced data, a model can score high accuracy by always predicting the majority class. If 1% of transactions are fraud, predicting “not fraud” every time is 99% accurate and useless. Precision, recall, PR-AUC or ROC-AUC give a better picture.
What is cross-validation?
Cross-validation splits the data into k folds, trains on k–1 folds and validates on the remaining one, rotating through all folds. It gives a more reliable estimate of performance than a single train–test split, especially with limited data.
Algorithms
How does a random forest work?
A random forest trains many decision trees on bootstrapped samples of the data, and each split considers only a random subset of features. Averaging (for regression) or voting (for classification) across trees reduces variance compared with a single tree.
What is gradient boosting?
Gradient boosting builds trees sequentially. Each new tree fits the errors of the current ensemble, gradually reducing the loss. It often performs very well on tabular data, but it needs careful tuning of learning rate and tree depth to avoid overfitting.
How do you handle missing data?
First understand why it’s missing. Options include dropping rows or columns with few values, imputing with the mean, median or a model, adding an indicator feature for “missing”, or using algorithms that handle missing values natively. The right choice depends on how much is missing and whether it’s random.
Practical and System Design
How would you build a recommendation system?
Clarify the goal and metrics (clicks, purchases, watch time). Start with a simple baseline such as popularity. Then use candidate generation (collaborative filtering or embeddings) to narrow millions of items to hundreds, and a ranking model to order them. Handle cold start for new users and items, and evaluate with offline metrics before running an A/B test.
How do you know a model is still working after deployment?
Monitor input data for drift, track prediction distributions and business metrics, and compare against ground-truth labels when they arrive. Set alerts and a retraining schedule, and keep the ability to roll back to a previous model.
Tell me about an ML project you worked on.
Structure it as problem, data, approach, evaluation and impact. Explain one decision you made and why, such as the metric you chose or a model you rejected, and what you would do differently.
How to Prepare
- Be ready to explain concepts simply, as if to a product manager.
- Practice coding basic data manipulation and a simple model from scratch.
- Prepare two or three project stories with measurable results.
- Review system design interview questions for senior roles.
Get Help in the Interview Itself
Preparation gets you most of the way. For the interview itself, ApplyQuick Interview Copilot listens to each question and suggests a structured answer based on your resume and the job description, on Zoom, Microsoft Teams, Google Meet or a phone call. You can also rehearse first with an AI mock interview. Every plan starts with a 7-day free trial.
