What should you focus on for a machine learning interview?
Can you explain how you would measure success?
Model evaluation was the most common topic in these interviews, appearing in 105 of 320 interviews. Questions went beyond naming a metric. They covered thresholds, calibration, fairness, and the quality of an evaluation.
Use these checks as you read:
- Can you explain precision and recall using a business example?
- Can you choose a metric for the problem and explain your choice?
- Can you explain how you would choose a prediction threshold?
- Can you describe how you would check a model for fairness?
Which basic ideas should you be able to explain?
The most asked question was about overfitting and underfitting, in 15 of 320 interviews. Learning types followed, in 14 of 320 interviews. Both questions asked for more than definitions: how to detect and prevent problems, or when to use each learning type.
- Can you explain how you would detect and prevent overfitting and underfitting?
- Can you say when to use supervised, unsupervised, self-supervised, or reinforcement learning?
- Can you explain the difference between AI, machine learning, and deep learning?
- Can you describe bagging and how it differs from boosting?
Can you connect your data choices to validation?
Data questions covered the prediction target, what each row represents, labels, missing values, and class imbalance. Validation questions covered data splits, preprocessing pipelines, time order, and grouped observations.
Can you state what your model should predict and what each row represents? Then explain how you would handle missing values, encode categories, and handle class imbalance. Can you split the data without leakage, including when observations are grouped or ordered by time?
In these interviews, the target and data granularity question was followed by a question about validation splits and leakage.
Can you explain what happens after deployment?
Deployment questions covered monitoring, changing data, retraining, security, and business impact. Use them to check more than how you would train a model.
- What would you check before releasing a model?
- What would you monitor after launch?
- How should a model adapt, or keep working with reduced capability, when its inputs or user behavior change?
- How would you report its business impact?
What code should you be ready to write?
Code-writing tasks appeared in 5 of 320 interviews. The recorded tasks offer concrete checks:
- Can you encode education levels or plan tiers as numeric values?
- Can you write Python code for logistic regression and get a yes/no prediction?
- Can you write linear regression code from loading data through making predictions?
Who gets asked Machine Learning
The 20 most asked Machine Learning questions
Counted by interview, not by mention: a question asked twice in one conversation counts once.
What are overfitting and underfitting, and how do you detect and prevent them?
Asked in 15 of 320 interviews · last asked Oct 23 · Concept · Junior
Show answer
Overfitting means learning training data too closely, including noise, and performing poorly on new data. Underfitting means missing useful patterns, often because the model is too simple. A large training-validation gap suggests overfitting; poor results on both suggest underfitting. Reduce overfitting with regularization or early stopping; reduce underfitting with better features, more model complexity, or enough training.
- Use a separate validation set or cross-validation to check performance on unseen data.
- Learning curves show how training and validation errors change with training time or data size.
- More representative training data can reduce overfitting; stronger regularization can worsen underfitting.
- Keep the test set untouched until final evaluation.
What the interviewer checks: The interviewer checks whether you can recognize each problem and choose changes that improve performance on unseen data.
Other ways it was asked
- “What are overfitting and underfitting?”
- “Describe overfitting, how you detect it, and how you prevent it.”
- “What is overfitting in machine learning?”
What are supervised, unsupervised, self-supervised, and reinforcement learning, and when would you use each?
Asked in 14 of 320 interviews · last asked Oct 19 · Concept · Junior
Show answer
Supervised learning uses labeled examples to predict outputs, such as spam labels. Unsupervised learning finds patterns without labels, such as customer groups. Self-supervised learning creates prediction targets from the data itself, often to pretrain models. Reinforcement learning learns actions through interaction and rewards, useful for sequential decisions such as robot control.
- Use supervised learning when you have labeled examples for the prediction task.
- Use unsupervised learning to explore structure, group similar items, or reduce dimensions.
- Use self-supervised learning to learn useful representations from unlabeled data, such as by predicting hidden words.
- Reinforcement learning aims to maximize total reward over time; rewards can be delayed.
What the interviewer checks: The interviewer checks whether you understand each learning signal and can match it to a suitable task.
Other ways it was asked
- “What is the difference between supervised and unsupervised learning?”
- “What is supervised learning?”
- “What are the types of machine learning?”
What are precision and recall, and how would you explain their differences using a business example?
Asked in 9 of 320 interviews · last asked Oct 16 · Concept · Junior
Show answer
Precision is the share of predicted positives that are truly positive. Recall is the share of actual positives the model finds. For customer attrition, precision measures how many flagged customers actually leave; recall measures how many leaving customers we flag. High precision limits wasted retention offers; high recall means fewer missed customers who leave.
- Precision = true positives / (true positives + false positives).
- Recall = true positives / (true positives + false negatives).
- Lowering the prediction threshold increases or maintains recall, but may reduce precision.
- Choose the threshold based on the costs of missed customers and unnecessary offers.
What the interviewer checks: The interviewer checks whether you can distinguish these metrics and connect them to business costs.
Other ways it was asked
- “How do precision and recall help evaluate a customer attrition model?”
- “What is the difference between precision and recall?”
- “What is precision and recall?”
How do you evaluate a machine learning model and choose appropriate metrics for the problem?
Asked in 8 of 320 interviews · last asked Oct 12 · Concept · Junior
Show answer
I evaluate a model on unseen data using metrics that match the task and the cost of mistakes. I use validation data or cross-validation to choose the model, then measure final performance on a separate test set. I compare it with a simple baseline and check whether it meets practical requirements.
- Choose data splits that prevent leakage; use time-based splits when predicting future events.
- For classification, use precision and recall when error costs differ; accuracy can mislead with imbalanced classes.
- For regression, use MAE for typical error size or RMSE to penalize large errors more.
- Check errors across important groups, prediction speed, and agreed performance targets before deployment.
What the interviewer checks: The interviewer wants to see whether you can test fairly and connect metric choices to real-world needs.
Other ways it was asked
- “How do you evaluate a model's performance?”
- “How do you determine whether a model is performing well or poorly, and what evaluation metrics do you look at?”
- “How do you evaluate a machine-learning model's accuracy?”
What is the difference between AI, machine learning, and deep learning?
Asked in 6 of 320 interviews · last asked Aug 9 · Concept · Junior
Show answer
AI is the broad field of building systems that perform tasks that usually need human intelligence. Machine learning is a part of AI where systems learn patterns from data instead of relying only on fixed rules. Deep learning is a part of machine learning that uses neural networks with many layers.
- AI includes both learning systems and rule-based systems.
- Deep learning often learns useful features directly from raw data, such as images.
- Deep learning often needs more data and computing power than simpler machine learning methods.
- Generative AI creates content, such as text or images; modern systems usually use deep learning.
What the interviewer checks: The interviewer checks whether you understand how these fields fit together and what makes each different.
Other ways it was asked
- “What is the difference between machine learning and deep learning?”
- “What is the difference between AI and deep learning?”
- “What's the difference between AI, machine learning, deep learning, and generative AI?”
How does bagging work, and how does it differ from boosting?
Asked in 6 of 320 interviews · last asked Oct 1 · Concept · Junior
Show answer
Bagging trains models independently on bootstrap samples: random samples drawn with replacement from the training data. It combines predictions by averaging or voting, mainly reducing variance. Boosting builds models in sequence, with each new model targeting errors in the current combined prediction. Boosting often reduces bias, but can be more sensitive to noise.
- Bagging models can train in parallel; boosting rounds usually depend on earlier rounds.
- Boosting may change example weights or fit loss gradients, depending on the method.
- Random forests use bagging plus random feature selection; AdaBoost and gradient boosting use boosting.
What the interviewer checks: The interviewer checks whether you understand how models are trained, combined, and used to reduce different errors.
Other ways it was asked
- “What is the difference between bagging and boosting?”
- “What is the basic difference between bagging and boosting?”
- “Could you describe the difference between boosting and bagging models?”
How would you choose the prediction target and data granularity for a machine learning model?
Asked in 5 of 320 interviews · last asked Sep 15 · Design · Mid-level
Show answer
I would choose a target that matches the decision the model will support and can be measured reliably. I would define exactly what to predict and how far ahead. I would choose what each row represents, such as a customer each day, based on when action is possible and what data is available.
- Define the outcome, prediction time, and time window for measuring the outcome.
- Match each row to the entity and time interval where a decision is made.
- Use only inputs available at prediction time to avoid data leakage.
- Finer detail can support specific actions but may leave too little data per group.
What the interviewer checks: The interviewer is checking whether you can turn a business goal into a clear, usable prediction task.
Other ways it was asked
- “How would you define the prediction target and data granularity for a machine learning system?”
- “How would you define the exact prediction target and data granularity for this system?”
- “How would you define the prediction target and data granularity for a predictive system?”
How do you handle class imbalance in a dataset?
Asked in 5 of 320 interviews · last asked Aug 11 · Scenario · Mid-level
Show answer
Class imbalance means some classes have far fewer examples than others. I compare class weights, oversampling rare classes, and undersampling common classes. I evaluate per-class precision and recall and macro F1, rather than accuracy alone. I choose the method and decision thresholds based on validation results and the cost of each error.
- Resample only training data, including within each cross-validation fold, to avoid data leakage.
- Keep validation and test class distributions representative of real use.
- Oversampling can increase overfitting; undersampling can remove useful information.
- Macro F1 gives each class equal weight, making it useful for multiclass imbalance.
What the interviewer checks: The interviewer checks whether you can improve rare-class predictions and evaluate them without misleading metrics or data leakage.
Other ways it was asked
- “How do you handle an imbalanced dataset?”
- “How would you handle an imbalanced dataset?”
- “How would you handle class imbalance in a predictive analysis problem with multiple classes?”
How would you structure your validation splits to prevent data leakage or look ahead bias?
Asked in 5 of 320 interviews · last asked Sep 11 · Scenario · Senior
Show answer
I would split data to match how the model will be used. For time-based predictions, I would train on earlier data and validate on later data. For predictions on new users or devices, I would keep each group in one split. I would fit preprocessing only on training data and keep a final test set untouched during tuning.
- Use rolling time splits to test performance across several future periods.
- Leave a time gap when training labels overlap the validation period.
- Use only features available at prediction time, not information added later.
- Fit imputation, scaling, and feature selection separately within each training fold.
What the interviewer checks: The interviewer checks whether your validation reflects real use and keeps future or held-out information out of training.
Other ways it was asked
- “How would you structure your validation splits to prevent data leakage or look-ahead bias?”
- “How would you structure validation splits to prevent data leakage or look-ahead bias?”
How should a deployed model adapt or gracefully degrade when its input data or user behavior changes?
Asked in 5 of 320 interviews · last asked Sep 14 · Scenario · Senior
Show answer
Monitor input data, user behavior, and model performance after deployment. If changes hurt quality, retrain on recent, representative data and test before a gradual rollout. While quality is uncertain, use safe rules, ask for human review, or stop making risky predictions. Keep rollback available if a new model performs worse.
- A change in input data does not always mean prediction quality has dropped.
- Track real outcomes when available; model confidence alone may miss failures.
- Validate updates before release, then monitor a small rollout before expanding.
- Define fallback triggers and safe behavior before problems happen.
What the interviewer checks: The interviewer is checking whether you can detect change, update safely, and protect users when predictions become unreliable.
Other ways it was asked
- “If data patterns or human behavior shift after deployment, how would you adapt the system or make it degrade gracefully?”
- “If underlying data patterns or human behaviors shift post deployment, how will your system adapt or gracefully degrade?”
- “How should a deployed predictive system adapt or gracefully degrade when data patterns or human behaviors shift?”
What is the difference between LLM and AI?
Asked in 4 of 320 interviews · last asked Oct 22 · Concept · Junior
Show answer
AI is the broad field of building systems that perform tasks requiring intelligence. An LLM, or large language model, is one type of AI trained to process and generate language. An AI agent is a system that takes actions toward a goal, often using an LLM to choose steps and tools.
- All LLMs are AI, but not all AI systems are LLMs.
- AI also includes systems for image recognition, recommendations, and other tasks.
- Agentic AI adds goal-directed actions, often through tools and feedback.
- An LLM alone is not automatically an AI agent.
What the interviewer checks: The interviewer is checking whether you can distinguish the broad field, a model type, and a system that takes actions.
Other ways it was asked
- “What is the difference between agentic AI and an LLM?”
- “What is the difference between an AI agent and an LLM?”
- “Is an LLM a form of AI?”
How would you handle missing values in training data?
Asked in 4 of 320 interviews · last asked Sep 23 · Scenario · Junior
Show answer
I would first check how much data is missing and why. Depending on the cause, I would drop affected rows or columns, fill missing values, or use a model that handles them. I would learn fill values from training data only and compare methods using validation data.
- Dropping data can lose useful information or introduce bias.
- Median fills are a simple numeric baseline; categorical values can use the mode or a missing category.
- Fit imputation within each training fold to avoid data leakage.
- Add a missing-value flag when missingness may help predict the target.
What the interviewer checks: The interviewer is checking whether you can handle missing data without causing bias or data leakage.
Other ways it was asked
- “How do you handle missing data?”
- “How do you deal with missing data?”
- “Your training data contains a large number of missing values. What would you do?”
What is XGBoost?
Asked in 4 of 320 interviews · last asked Sep 19 · Concept · Junior
Show answer
XGBoost stands for Extreme Gradient Boosting. It is a machine learning library commonly used to build boosted decision trees. It adds trees one at a time, using loss gradients to reduce errors in the current model. Regularization, learning rate, and tree limits help control overfitting.
- Commonly used for classification and regression on tabular data.
- Uses first and second derivatives of the loss to guide tree building.
- Supports missing values and efficient training.
- Can still overfit, so tune model settings and use validation data for early stopping.
What the interviewer checks: The interviewer checks whether you understand how XGBoost builds trees sequentially and controls overfitting.
Other ways it was asked
- “What is XGBoost, and how does it work?”
- “How does XGBoost work?”
- “Explain XGBoost.”
What is the difference between L1 and L2 regularization?
Asked in 4 of 320 interviews · last asked Aug 11 · Concept · Junior
Show answer
L1 regularization adds a penalty based on the sum of absolute weight values; L2 uses the sum of squared weights. Both help reduce overfitting by discouraging large weights. L1 can make weights exactly zero, which helps select features. L2 usually shrinks weights without making them exactly zero.
- L1 encourages sparse models: some features get zero weights.
- L2 spreads shrinkage across weights and can help when features are strongly correlated.
- A larger regularization strength means a stronger penalty.
- Scale features so their units do not unfairly affect the penalty.
What the interviewer checks: The interviewer is checking whether you understand how each penalty affects weights, feature selection, and overfitting.
Other ways it was asked
- “What is L2 regularization?”
- “What are L1 and L2 regularization?”
- “What are L1 and L2 regularization, and what are they used for?”
What is gradient descent?
Asked in 4 of 320 interviews · last asked Sep 4 · Concept · Junior
Show answer
Gradient descent is a method for reducing a model’s loss by repeatedly adjusting its parameters. It calculates the loss gradient and moves the parameters in the opposite direction. The learning rate scales each step, while the gradient determines its direction and also affects its size.
- The gradient points toward the steepest local increase in loss.
- Each update subtracts the learning rate times the gradient from the parameters.
- A large learning rate can cause overshooting; a small one can make training slow.
- Gradient descent does not guarantee finding the global minimum.
What the interviewer checks: The interviewer checks whether you understand how gradients guide parameter updates and how the learning rate affects training.
Other ways it was asked
- “What do you mean by gradient descent?”
- “What is gradient descent, how does it work, and what determines its pace?”
- “What happens under the hood in gradient descent?”
What are the differences between training, validation, and test data?
Asked in 4 of 320 interviews · last asked Sep 15 · Concept · Junior
Show answer
Training data is used to learn the model’s parameters. Validation data is used to tune settings and choose between models. Test data is kept separate until those choices are final, then used to estimate performance on unseen data.
- Training changes learned parameters; validation guides choices such as hyperparameters and when to stop training.
- Using test results to tune the model makes the final performance estimate less trustworthy.
- Fit preprocessing steps on training data only, then apply them to validation and test data.
- Avoid duplicate records across splits; separate users and preserve time order when the task requires it.
What the interviewer checks: The interviewer checks whether you understand each split’s role and how to avoid data leakage.
Other ways it was asked
- “What is the difference between training data and test data?”
- “What is the difference between training data and testing data?”
- “What is the difference between test data and validation data?”
How would you evaluate a severely imbalanced classifier without being misled by accuracy?
Asked in 4 of 320 interviews · last asked Oct 22 · Concept · Mid-level
Show answer
I prioritize precision, recall, and the precision-recall curve for the minority class, rather than accuracy. I summarize that curve with average precision and inspect the confusion matrix at the chosen threshold. I choose the threshold on validation data using error costs. I report final results on untouched test data with realistic class proportions and no leakage.
- High accuracy can hide a model that never detects the minority class.
- Precision measures how often positive predictions are correct; recall measures how many actual positives are found.
- F1 balances precision and recall, but does not capture unequal error costs.
- ROC-AUC can look strong despite poor precision; precision-recall results depend on class proportions.
What the interviewer checks: The interviewer checks whether you can measure minority-class performance and choose a realistic, leakage-free evaluation setup.
Other ways it was asked
- “When evaluating a binary classifier on imbalanced data, what metrics do you prioritize, and why?”
- “When evaluating a binary classifier with strong class imbalance, which metrics would you rely on most and why?”
- “What evaluation setup and metrics do you use to avoid being misled by accuracy on a severely imbalanced dataset?”
Can you explain a confusion matrix?
Asked in 4 of 320 interviews · last asked Oct 9 · Concept · Junior
Show answer
A confusion matrix is a table that compares predicted classes with actual classes. For binary classification, it shows true positives, true negatives, false positives, and false negatives. It helps us see which mistakes a model makes, not just how often it is correct.
- True positives and true negatives are correct predictions of the positive and negative classes.
- False positives are false alarms; false negatives are missed positive cases.
- Check axis labels: rows and columns may represent actual or predicted classes.
- For multiple classes, each cell counts cases with one actual class and one predicted class.
CodePredicted positive Predicted negative Actual positive 40 true positives 10 false negatives Actual negative 5 false positives 45 true negativesWhat the interviewer checks: The interviewer wants to know whether you can identify correct predictions and distinguish the types of classification errors.
Other ways it was asked
- “What is a confusion matrix?”
- “What do you understand by a confusion matrix?”
- “What are the components of a confusion matrix?”
What are embeddings?
Asked in 4 of 320 interviews · last asked Oct 19 · Concept · Junior
Show answer
Embeddings are vectors of numbers that represent items such as words, images, or users. They are usually learned from data, so items with similar meaning or behavior have nearby vectors. They are used in semantic search, recommendation systems, and as inputs to machine learning models.
- Embeddings capture useful relationships, not just unique labels for items.
- Similarity can be measured with cosine similarity, dot product, or distance, depending on the model.
- What counts as similar depends on the training data and task.
CodeA search model embeds a question and a set of documents. It compares their vectors to find relevant documents. This can find matches even when the exact words differ.What the interviewer checks: The interviewer is checking whether you understand what embeddings represent and how they support similarity-based tasks.
Other ways it was asked
- “What is your understanding of embeddings?”
- “What are embeddings, and when are they used?”
- “What are embeddings, and where are they used?”
How would you explain machine learning to a high school student?
Asked in 3 of 320 interviews · last asked Aug 13 · Concept · Junior
Show answer
Machine learning is a way for computers to learn patterns from examples instead of being given a rule for every case. For example, a spam filter learns from emails labeled as spam or not spam, then predicts whether new emails are spam. It can make mistakes, especially when new examples differ from what it learned from.
- Training means learning patterns from data.
- The goal is to make useful predictions on new examples, not just memorize old ones.
- Poor or unrepresentative training data can lead to poor predictions.
What the interviewer checks: The interviewer wants to see whether you can explain the core idea accurately without technical terms.
Other ways it was asked
- “What is machine learning?”
- “How would you explain machine learning to a 9th or 10th grade student?”
Machine learning fundamentals interview questions
Check how you explain machine learning, learning types, rule-based alternatives, basic linear algebra, and Python tools. 74 questions.
How would a traditional rule-based solution to distinguish apples from oranges compare with a machine learning-based solution?
Asked in 2 of 320 interviews · last asked May 7 · Concept · Junior
Other ways it was asked
- “How would you implement a program to distinguish apples from oranges using traditional methods, and how would you approach it with machine learning?”
Can you tell me what AI is?
Asked in 2 of 320 interviews · last asked Oct 3 · Concept · Junior
Other ways it was asked
- “What is an AI?”
Can you explain what generative AI is?
Asked in 2 of 320 interviews · last asked Jul 31 · Concept · Junior
Other ways it was asked
- “What is generative AI, and what does it do?”
Can AI systems be built using open-source or closed-source models, and how do those approaches differ?
Asked in 2 of 320 interviews · last asked Oct 13 · Concept · Junior
Other ways it was asked
- “What is the difference between open-source and closed-source AI models?”
Why is Python used in machine learning?
Asked in 2 of 320 interviews · last asked Aug 14 · Concept · Junior
Other ways it was asked
- “Why is Python used in AI and data science?”
What are tensors, and how are they used in machine learning?
Asked once · last asked Jul 16 · Concept · Junior
Given two matrix shapes, can you multiply them, and what are the output dimensions?
Asked once · last asked Jul 16 · Concept · Junior
Other ways it was asked
- “If A is 128×256 and B is 256×800, what is the output dimension when you multiply them?”
- “If A is 128×256 and B is 800×246, what is the output matrix?”
- “Can a 2×3 matrix be multiplied by a 3×2 matrix, and what are the dimensions of the result?”
What data do you pass to scikit-learn to train a model, and what does the training call return?
Asked once · last asked Sep 10 · Concept · Junior
What is supervised machine learning, and how are training and test data used?
Asked once · last asked Mar 24 · Concept · Junior
What type of machine learning would you use to train a robot to walk on newly discovered planets, with no prior training data, learning from each new terrain it encounters?
Asked once · last asked Apr 20 · Scenario · Mid-level
What is data mining, and how does it differ from simple data querying?
Asked once · last asked May 1 · Concept · Junior
What are the major data mining functionalities?
Asked once · last asked May 1 · Concept · Junior
Data preprocessing and feature engineering interview questions
Questions cover choosing a target, defining rows, labeling data, handling missing values and imbalance, and selecting or encoding features. 99 questions.
What is feature engineering, and how does it differ from feature selection?
Asked in 3 of 320 interviews · last asked Aug 27 · Concept · Junior
Other ways it was asked
- “What is feature engineering?”
- “How would you do feature selection and feature engineering?”
- “What is the difference between feature engineering and feature selection?”
How do you measure whether feature engineering changes improve a machine-learning model?
Asked in 2 of 320 interviews · last asked Oct 16 · Concept · Mid-level
Other ways it was asked
- “Why can feature engineering improve a machine-learning model's performance?”
Regarding supervised learning, how do you handle situations where labeled data is scarce?
Asked in 2 of 320 interviews · last asked Oct 17 · Scenario · Mid-level
Other ways it was asked
- “What would you do if labeled data were scarce for a machine learning problem?”
What is Synthetic Minority Oversampling Technique (SMOTE)?
Asked once · last asked Oct 1 · Concept · Junior
Other ways it was asked
- “What does Synthetic Minority Oversampling Technique, or SMOTE, do?”
How would you define customer churn and construct positive labels for a churn prediction model?
Asked once · last asked Oct 20 · Case · Mid-level
Other ways it was asked
- “What criteria define a customer as churned?”
- “How would you define a customer as churned for a churn analysis?”
- “How would you construct positive labels for churn prediction?”
How do you construct the dataset for churn prediction?
Asked once · last asked Oct 20 · Case · Mid-level
Other ways it was asked
- “How would you construct training data for a churn prediction model?”
Before building a model, we generally preprocess the data. What are some things you have to look out for?
Asked once · last asked Mar 5 · Concept · Mid-level
What criteria do you use to optimize features for a machine learning model?
Asked once · last asked Mar 5 · Concept · Mid-level
What are some common approaches to reducing the number of features?
Asked once · last asked Mar 5 · Concept · Mid-level
What features would you consider in a bank fraud detection system to prevent overfitting?
Asked once · last asked Mar 13 · Scenario · Mid-level
How can ANOVA be used for feature selection?
Asked once · last asked Apr 9 · Concept · Junior
How would you handle missing data when the hotel segments are imbalanced, with many rows for some segments and very few for others?
Asked once · last asked Apr 13 · Scenario · Mid-level
Other ways it was asked
- “How would you impute missing hotel ratings when ratings are an important feature and the hotel segments are imbalanced?”
Machine learning algorithms and model selection interview questions
Compare regression, classification, ensembles, clustering, PCA, and association rules, then explain model choices and predictions. 101 questions.
How does a random forest work?
Asked in 3 of 320 interviews · last asked Oct 1 · Concept · Junior
Other ways it was asked
- “What is a random forest?”
What is logistic regression?
Asked in 3 of 320 interviews · last asked Aug 27 · Concept · Junior
Other ways it was asked
- “What is logistic regression, and how does it work?”
- “Can you explain logistic regression?”
What are some common supervised and unsupervised machine learning algorithms?
Asked in 3 of 320 interviews · last asked Aug 21 · Concept · Junior
Other ways it was asked
- “Can you name a few commonly used machine learning algorithms?”
- “What are examples of algorithms used for supervised and unsupervised learning?”
- “What are some common algorithms used in supervised and unsupervised learning, and how do they differ?”
How do you choose the right machine learning model for a problem and assess its fit?
Asked in 2 of 320 interviews · last asked Aug 22 · Concept · Mid-level
Other ways it was asked
- “How do you choose the right ML algorithm for a problem?”
- “How do you choose the right machine-learning algorithm for a problem?”
- “Which models would you choose for an AI application, and how would you evaluate their fit?”
What type of algorithm is K-Nearest Neighbors (KNN)?
Asked in 2 of 320 interviews · last asked Aug 13 · Concept · Junior
Other ways it was asked
- “How does the K-nearest neighbors (KNN) algorithm work, and when is it useful?”
When choosing between models, how do you assess whether the simpler model's interpretability is worth potentially sacrificing some accuracy?
Asked in 2 of 320 interviews · last asked Aug 31 · Scenario · Mid-level
Other ways it was asked
- “If you had to choose between deploying a highly accurate but complex model and a slightly less accurate but more interpretable one, how would you make that decision?”
- “How would you choose between two predictive models, one with higher accuracy but lower interpretability and the other the opposite?”
What roles do support and confidence play in association rule mining?
Asked once · last asked May 2 · Concept · Mid-level
Other ways it was asked
- “How are support and confidence used in association rule mining?”
- “What is all-confidence in association rule mining?”
What types of constraints are used in association rule mining?
Asked once · last asked May 2 · Concept · Mid-level
How are principal components ranked in PCA?
Asked once · last asked May 25 · Concept · Mid-level
Other ways it was asked
- “Why are principal components ranked that way in PCA?”
How do you decide whether to use traditional machine learning or deep learning for a problem?
Asked once · last asked Sep 22 · Concept · Senior
How do support vector machines (SVMs) classify data?
Asked once · last asked Apr 20 · Concept · Junior
In what scenarios would you use an SVM model?
Asked once · last asked Apr 20 · Concept · Junior
All 101 Machine learning algorithms and model selection questions
Model training and hyperparameter tuning interview questions
Work through loss functions, learning rates, regularization, neural networks, transfer learning, and fine-tuning. 68 questions.
What are bias and variance, how do they relate to RMSE, and what is their tradeoff?
Asked in 3 of 320 interviews · last asked Sep 7 · Concept · Junior
Other ways it was asked
- “Tell me about the bias-variance trade-off and how bias and variance are related to RMSE.”
- “How are bias and variance defined and calculated, and how are they related to RMSE?”
- “What is the equation relating RMSE, bias, and variance?”
How would you improve the accuracy of a predictive model?
Asked in 2 of 320 interviews · last asked Mar 27 · Scenario · Mid-level
Other ways it was asked
- “How would you improve a machine learning model?”
Describe some common machine learning architectures and when you would use them.
Asked in 2 of 320 interviews · last asked Oct 19 · Concept · Mid-level
Other ways it was asked
- “Describe common machine learning architectures, their typical use cases, and briefly note any limitations.”
How do you train a machine-learning model?
Asked in 2 of 320 interviews · last asked Sep 10 · Concept · Junior
Other ways it was asked
- “How would you train a machine learning model?”
What is the learning rate, and how does increasing or decreasing it affect training?
Asked in 2 of 320 interviews · last asked Oct 9 · Concept · Mid-level
Other ways it was asked
- “Can you explain learning rate in simple language?”
- “How do the trade-offs between a low and a high learning rate affect model training?”
- “What does the learning rate mean in machine learning?”
What does it mean to freeze the model head during transfer learning?
Asked once · last asked Sep 16 · Concept · Mid-level
Other ways it was asked
- “When should you freeze the model head during transfer learning?”
How would you decide whether to spend limited time collecting more data or improving the model’s algorithm to achieve the best results?
Asked once · last asked Oct 20 · Scenario · Senior
Other ways it was asked
- “How would you decide whether to collect more data or improve a machine learning algorithm when time is limited?”
What initial parameters would you use to fit a linear regression model to data in three-dimensional space, and how would you optimize them?
Asked once · last asked Mar 4 · Concept · Junior
Could you explain how structured feedback can optimize the performance of machine learning models?
Asked once · last asked Apr 8 · Concept · Mid-level
How do data augmentation and transfer learning help when labeled data is scarce in supervised learning?
Asked once · last asked Apr 21 · Concept · Mid-level
What is hyperparameter tuning, and when would you use it?
Asked once · last asked May 22 · Concept · Mid-level
What strategies would you use for hyperparameter tuning in object detection models?
Asked once · last asked May 22 · Scenario · Mid-level
Cross-validation and data leakage interview questions
Explain how data splits, preprocessing pipelines, time order, and grouped observations affect leakage and reliable testing. 41 questions.
What validation techniques can you use to evaluate a model, and how does cross-validation work?
Asked in 3 of 320 interviews · last asked Aug 21 · Concept · Junior
Other ways it was asked
- “Which validation techniques can you use to evaluate a machine-learning model?”
- “What coding techniques or other methods would you use to validate a machine-learning model?”
- “Can you describe what cross-validation is?”
What is data leakage, and how would you identify and prevent it in a modeling workflow?
Asked in 3 of 320 interviews · last asked Oct 18 · Scenario · Mid-level
Other ways it was asked
- “How do you identify data leakage in a proposed modeling workflow?”
- “What is data leakage in machine learning? Give a real-world example and explain how you would prevent it.”
- “How would you prevent data leakage when building a machine learning model?”
In scikit-learn, how do you correctly run cross-validation when preprocessing steps such as scaling and one-hot encoding are involved, to avoid data leakage?
Asked in 2 of 320 interviews · last asked Jun 3 · Tooling · Mid-level
Other ways it was asked
- “In scikit-learn, how do you avoid data leakage when doing preprocessing and cross-validation, especially with scaling and imputation?”
How do you validate transaction cost and slippage assumptions in a rates backtest against realized fills?
Asked once · last asked Oct 10 · Scenario · Senior
Other ways it was asked
- “How do you model transaction costs and slippage in a rates backtest, such as for U.S. Treasury futures or swaps, and validate those assumptions against realized fills?”
What data would you use to validate a campus placement prediction model?
Asked once · last asked Mar 27 · Scenario · Mid-level
How do you design validation when there's a risk of leakage, such as with time-series data, multiple rows per user, or repeated interactions in a recommender system?
Asked once · last asked May 3 · Scenario · Senior
In scikit-learn, how do you prevent target leakage when doing time-series cross-validation with feature scaling and encoding, and what APIs do you use to enforce that?
Asked once · last asked May 3 · Tooling · Senior
How would you structure a scikit-learn pipeline that includes lag or rolling features from a timestamped series without leaking future information, given that a pipeline doesn't natively know about time order?
Asked once · last asked May 3 · Scenario · Senior
In a custom transformer, how would you handle the boundary between training and validation so the first validation rows can use the last k training observations for lag or rolling features without refitting on validation targets?
Asked once · last asked May 3 · Scenario · Senior
How do you prevent data leakage when doing target encoding for a high-cardinality categorical feature, especially under time-based cross-validation?
Asked once · last asked May 3 · Scenario · Senior
When splitting a dataset into groups, should the groups have similar data distributions or equal numbers of samples?
Asked once · last asked May 7 · Concept · Junior
When you use GridSearchCV with a pipeline and a ColumnTransformer, where should fitting happen, and what data should never be used to learn preprocessing parameters?
Asked once · last asked Jun 3 · Tooling · Senior
Model evaluation metrics interview questions
Choose and explain metrics, thresholds, calibration, uncertainty, benchmarks, and fairness checks for model evaluation. 161 questions.
Which metrics would you monitor during AI model training?
Asked in 3 of 320 interviews · last asked Oct 2 · Concept · Mid-level
Other ways it was asked
- “Which metrics do you monitor when benchmarking AI models?”
- “Which metrics would you check during machine learning model training?”
How would you measure whether fine-tuning improved the model?
Asked in 2 of 320 interviews · last asked Sep 27 · Scenario · Mid-level
Other ways it was asked
- “How would you determine whether fine-tuning actually improves a model?”
When would you prioritize precision over recall, and what are the limitations of confusion-matrix metrics?
Asked in 2 of 320 interviews · last asked Sep 9 · Concept · Mid-level
Other ways it was asked
- “When would you prefer precision over recall?”
- “When would you prefer recall over precision?”
- “Why might precision be preferred over recall, and what are the limitations of relying on confusion-matrix metrics?”
What is the difference between F1 and ROC-AUC?
Asked in 2 of 320 interviews · last asked Aug 12 · Concept · Mid-level
Other ways it was asked
- “What are the ROC curve and the AUC metric, and how do they differ?”
If your dataset is 99:1 imbalanced, which metrics would you choose for the model, and how would you set the decision threshold for deployment?
Asked in 2 of 320 interviews · last asked Jun 30 · Scenario · Mid-level
Other ways it was asked
- “When you evaluate a binary classification model for an imbalanced dataset, which metrics do you prioritize, and how do you choose the decision threshold for production?”
How would you verify a model evaluation answer involving metrics like accuracy, precision, recall, F1, ROC AUC, or RMSE?
Asked in 2 of 320 interviews · last asked Jun 26 · Scenario · Mid-level
Other ways it was asked
- “When evaluating model performance, how do you assess it beyond a single metric, and when might you favor precision, recall, F1, or ROC AUC?”
What evaluation data, tasks, and metrics would you use to measure a face recognition system's accuracy?
Asked in 2 of 320 interviews · last asked Sep 22 · Scenario · Mid-level
Other ways it was asked
- “How should you measure and interpret the accuracy of a face-recognition system such as DeepFace?”
- “What data or evaluation task did you use to measure a face recognition model's accuracy?”
What is an F1 score?
Asked in 2 of 320 interviews · last asked Sep 29 · Concept · Junior
Other ways it was asked
- “What is the F1 score?”
How would you evaluate the performance of an object-detection model?
Asked once · last asked Sep 7 · Scenario · Mid-level
Other ways it was asked
- “Which metrics would you use to evaluate an object-detection model?”
How would you evaluate AI or data science work using a structured rubric?
Asked once · last asked Oct 14 · Behavioral · Mid-level
Other ways it was asked
- “Tell me about a recent example of evaluating AI or data science work using a structured rubric.”
How can precision be calculated for supervised models, and how does evaluating precision differ for unsupervised models?
Asked once · last asked Mar 5 · Concept · Mid-level
How do you process data and evaluate a machine-learning model using accuracy, precision, recall, and a confusion matrix?
Asked once · last asked Mar 24 · Concept · Junior
Applied machine learning interview questions
Explore complete machine learning solutions for vision, text, audio, generative AI, forecasting, robotics, healthcare, finance, and industrial problems. 79 questions.
How does a hate-speech detection model work technically?
Asked once · last asked May 18 · Concept · Mid-level
Other ways it was asked
- “What does a hate-speech detection model do?”
- “How does a hate-speech detection model work, and what is its process?”
How would you choose between FaceNet and other face recognition models, considering accuracy and deployment constraints?
Asked once · last asked Sep 22 · Scenario · Mid-level
Other ways it was asked
- “What benchmark metrics would you compare when selecting FaceNet over other face recognition models?”
- “What criteria would you use to select FaceNet over other face recognition models?”
- “Which metrics and deployment factors, such as latency, model size, false positives, false negatives, precision, and recall, matter when selecting a face recognition model?”
What are the dominant sources of false positives in lunar impact-flash detection?
Asked once · last asked Apr 15 · Scenario · Mid-level
Other ways it was asked
- “How can a model distinguish false positives in lunar impact-flash detection?”
Could a lunar impact-flash detection pipeline be generalized to detect space-based events such as glints or debris?
Asked once · last asked Apr 15 · Scenario · Senior
Other ways it was asked
- “What changes would you make to adapt a lunar impact-flash detection pipeline to detect space-based events?”
How would you handle variation in screw locations across supplier images when building an image-segmentation model?
Asked once · last asked Jul 21 · Scenario · Senior
Other ways it was asked
- “Would you still use an image-segmentation model when screw locations vary across supplier images?”
How would you start building an ML pipeline to classify reviews and call transcripts by sentiment, and what approach would you take first?
Asked once · last asked Jul 27 · Design · Mid-level
Other ways it was asked
- “How would you start building an ML pipeline to classify reviews and call transcripts?”
How would you solve the N-Queens problem using reinforcement learning?
Asked once · last asked Aug 13 · Scenario · Mid-level
What parameters would you consider when comparing MobileNet SSD and YOLO?
Asked once · last asked Sep 16 · Concept · Mid-level
Other ways it was asked
- “What parameters would you consider when choosing between MobileNet-SSD and YOLO?”
How would you choose a model for detecting defects in fabric?
Asked once · last asked Oct 21 · Scenario · Mid-level
Other ways it was asked
- “How would you validate a model that detects defects in fabric?”
- “What metrics would you use to evaluate a fabric defect-detection model?”
How would you build and evaluate an AI-based URL detector for phishing emails, including measuring its accuracy?
Asked once · last asked Mar 26 · Scenario · Mid-level
Can you describe a specific project where you used Python and machine learning frameworks to solve a real-world problem, including the problem, your approach, and the outcome?
Asked once · last asked Mar 27 · Behavioral · Mid-level
How would you design a campus placement analytics and performance prediction system?
Asked once · last asked Mar 27 · Design · Mid-level
Search and recommendation system interview questions
These questions connect embeddings and similarity with retrieval, RAG, ranking, recommendations, cold starts, and relevance. 41 questions.
When would you use supervised rather than unsupervised learning for a recommendation model?
Asked once · last asked Sep 28 · Concept · Mid-level
Other ways it was asked
- “How would you use supervised learning to train a recommendation model?”
- “Why might supervised learning be a better choice than unsupervised learning for a recommendation model?”
- “How do supervised learning and reinforcement learning differ when building a recommendation model?”
Which library would you use to calculate cosine similarity and match items?
Asked once · last asked Aug 19 · Tooling · Mid-level
How would you evaluate a hybrid BM25 and k-nearest-neighbor vector-search approach, and which metrics would you use?
Asked once · last asked Sep 7 · Concept · Senior
Other ways it was asked
- “How would you test whether adding k-nearest-neighbor vector search to BM25 improves performance compared with a BM25 baseline?”
- “How would you evaluate a hybrid BM25 and vector-search approach, and how would you choose the weights for each retrieval method?”
How do embeddings work in a computer system, and why are they used?
Asked once · last asked Apr 16 · Concept · Junior
Comparing the last 90 days of recommendations with customer behavior, how would you define success and a reported 30% failure rate precisely enough to identify patterns and verify that rule changes fixed the problem rather than shifted it elsewhere?
Asked once · last asked May 7 · Scenario · Senior
Other ways it was asked
- “When you compare the last 90 days of recommendations with customer behavior, what exactly are you measuring as success? How do you define the 30% failure rate precisely enough to identify a pattern, and how do you know your rule changes fixed it rather than moving the problem somewhere else?”
What data would you use to build an offer-to-user ranking model, and how would you distinguish available features from assumptions?
Asked once · last asked May 13 · Scenario · Senior
Which retrieval metrics can help assess whether retrieved chunks are complete and relevant?
Asked once · last asked May 13 · Concept · Mid-level
How would you evaluate a model's accuracy and retrieval quality using metrics such as precision, recall, or mean average precision?
Asked once · last asked Jun 12 · Scenario · Senior
How would you design an offline evaluation that reliably predicts the online impact of a ranking or recommendation model, including how you would avoid leakage and handle delayed feedback?
Asked once · last asked Jun 15 · Scenario · Senior
How do you compute and apply propensity scores for the behavior policy in inverse propensity scoring, and how do you control variance from very small propensities?
Asked once · last asked Jun 15 · Concept · Senior
How would you estimate the true relevance of items from biased click logs, for example, by accounting for position bias in an offline evaluation?
Asked once · last asked Jun 15 · Scenario · Senior
How would you handle variance and exploding weights in inverse propensity scoring, for example, when propensities are very small at low ranks, while keeping the estimator reliable?
Asked once · last asked Jun 15 · Concept · Senior
MLOps and model deployment interview questions
Discuss production pipelines, launch checks, monitoring, changing data, retraining, security, and how to report business impact. 79 questions.
How do you mathematically handle live data that's highly sparse, imbalanced, or unrepresentative of the environment your model encounters post launch?
Asked in 3 of 320 interviews · last asked Sep 11 · Scenario · Senior
Other ways it was asked
- “How would you mathematically handle live data that is sparse, balanced, or unrepresentative of the environment a model encounters after launch?”
- “What methods would you use to handle live data that is sparse, imbalanced, or unrepresentative of a model's post-launch environment?”
How would you distinguish data drift, concept drift, and temporary data issues when production performance drops?
Asked in 3 of 320 interviews · last asked Aug 19 · Scenario · Senior
Other ways it was asked
- “In a production ML system, how would you detect and handle data drift versus concept drift, and what metrics or signals would you monitor for each?”
- “Suppose your ML model's accuracy drops from 91% to 78% in production. How would you determine whether this is data drift, concept drift, or a temporary data issue?”
- “In production, how would you detect and respond to data drift versus concept drift for a deployed classifier?”
How would you deploy a machine learning model for business intelligence or analytics engineering, and how would it be used?
Asked in 2 of 320 interviews · last asked Aug 31 · Behavioral · Senior
Other ways it was asked
- “Describe a machine learning model you personally deployed to production within your area of specialization.”
- “Tell me about a machine learning model you deployed in your area of specialization and how it was used.”
How would you investigate a model that performs well during development but poorly in production?
Asked in 2 of 320 interviews · last asked Sep 23 · Scenario · Mid-level
Other ways it was asked
- “How would you debug a model that performs well in training but poorly in production?”
- “How do you handle a model that performs well during development but poorly during production?”
- “Your ML model has 95% accuracy in testing but only 75% in production. What would you do?”
What causes data drift?
Asked in 2 of 320 interviews · last asked Sep 7 · Concept · Mid-level
Other ways it was asked
- “What is data drift?”
How would you train a machine learning model using Xcode or Python scripts, and integrate it with Core ML?
Asked once · last asked Feb 27 · Tooling · Mid-level
How would you communicate with a business unit to gather the information needed and assess whether a model is working effectively?
Asked once · last asked Mar 12 · Scenario · Mid-level
How would you identify and address failure modes in a machine-learning project?
Asked once · last asked Apr 15 · Scenario · Senior
Besides accuracy, what signs could indicate that a machine-learning model is underperforming or something is going wrong?
Asked once · last asked Apr 15 · Scenario · Mid-level
What's your typical approach to packaging and serving a model in batch versus real time, and how do you ensure training-serving feature parity?
Asked once · last asked May 3 · Design · Senior
Once a model is in production, what monitoring do you set up for data drift, performance, calibration, and latency, and what thresholds or triggers cause you to retrain or roll back?
Asked once · last asked May 3 · Scenario · Senior
How would you monitor a model when ground-truth labels arrive with a long delay or are partially missing, including your choice of proxies and evaluation setup?
Asked once · last asked May 3 · Scenario · Senior
Other ways it was asked
- “How do you handle monitoring when ground-truth labels arrive with a long delay or are partially missing? What proxies or evaluation setup do you use?”
Related interview questions
- Data Scientist interview questions
- AI Engineer interview questions
- Machine Learning Engineer interview questions
- AI Trainer interview questions
- Software Engineer interview questions
- Data Analyst interview questions
- Data Engineer interview questions
- Deep Learning interview questions
- Computer Vision interview questions
- Generative AI and LLM interview questions
- RAG interview questions
- Agentic AI interview questions
- MLOps interview questions
How this list is made
Every question here was asked by a real interviewer in a real job interview that a candidate ran MeetAssist in. After the call, the interviewer's questions are pulled from the transcript. Practice sessions, tests, exams and anything that was not a job interview are dropped.
Questions are rewritten so they make sense on their own: names of people, employers and clients are removed, and speech-to-text mistakes are fixed. Small talk, logistics and questions about one candidate's own résumé are left out. Rephrasings of the same question are merged, and the count is the number of different interviews it came up in.
The interviews come from candidates who use MeetAssist, so they lean towards software, data and AI roles: the shares describe these interviews, not every Machine Learning interview. Answers to the most asked questions are written and checked by an editor; a "common mistake" is shown only when reviews of real answers in these interviews show it, with the count.
320 interviews, 221 candidates, Nov 2025 to Oct 2026. Updated every month with new interviews.
Or add the Chrome extension