All 99 questions
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 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 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?”
You have 1,000 rows of hotel data, and 50 rows are missing the hotel rating column. How would you handle that?
Asked once · last asked Apr 13 · Scenario · Mid-level
How would you choose or create a dataset for training a prediction model?
Asked once · last asked May 22 · Scenario · Mid-level
How would you address spatial resolution challenges and class imbalance when working with satellite imagery?
Asked once · last asked May 22 · Scenario · Mid-level
When using a ColumnTransformer with OneHotEncoder, how do you handle unseen categories at inference time, and what trade-offs does your choice introduce?
Asked once · last asked May 24 · Tooling · Mid-level
In scikit-learn's OneHotEncoder, what's the practical difference between handle_unknown="ignore" and handle_unknown="infrequent_if_exist", and when would you prefer each?
Asked once · last asked May 24 · Concept · Mid-level
How would you source training and test datasets for a machine-learning project?
Asked once · last asked Jun 9 · Scenario · Mid-level
How do you typically handle missing data and outliers in a data set, and what factors guide your choice of technique?
Asked once · last asked Jun 26 · Scenario · Mid-level
Which predictive imputation techniques are available, and how do you choose among them?
Asked once · last asked Jun 26 · Concept · Mid-level
When building a binary classification model with severe class imbalance, which techniques would you consider and why?
Asked once · last asked Jun 26 · Concept · Mid-level
How would you ensure data integrity and engineer robust features when integrating data from sources with varying data quality and schema inconsistencies for a predictive maintenance model?
Asked once · last asked Jul 16 · Scenario · Senior
How would you design your data preparation and feature engineering pipeline for a real-time credit scoring model with low feature latency, balancing accuracy and efficiency?
Asked once · last asked Jul 16 · Design · Senior
What are common methods for feature selection?
Asked once · last asked Jul 15 · Concept · Mid-level
Given the trend toward making evaluation data distribution-aware, how do you see this impacting strategies for training data collection and preprocessing in the future?
Asked once · last asked Jul 23 · Senior
How would you perform feature engineering for a skin cancer detection application?
Asked once · last asked Jul 27 · Scenario · Mid-level
When auditing a labeled dataset for model training, what sampling strategy would you use to detect systematic bias or label leakage, and what evidence would confirm the issue?
Asked once · last asked Aug 4 · Scenario · Senior
What is the difference between standardization and normalization, and why are they used?
Asked once · last asked Aug 11 · Concept · Junior
When you start a new ML project, how do you decide whether to prioritize improving data quality or labels versus changing the model architecture?
Asked once · last asked Aug 11 · Scenario · Mid-level
When cleaning a very large dataset for AI training, what core validation checks would you run to catch duplicates, missing values, outliers, and label leakage before approving the data?
Asked once · last asked Aug 19 · Scenario · Senior
How do feature engineering and data processing support machine-learning workflows?
Asked once · last asked Aug 19 · Concept · Mid-level
How do you handle missing values, outliers, and class imbalance in a machine-learning dataset?
Asked once · last asked Aug 22 · Scenario · Mid-level
When annotating multimodal data, how do you decide whether labels should be mutually exclusive or allow multiple labels? What is an example of how the wrong choice can harm model performance?
Asked once · last asked Aug 22 · Concept · Senior
When you're preparing a dataset for an ML model, what steps do you take to detect and treat outliers and missing values without causing train-serve skew?
Asked once · last asked Aug 23 · Scenario · Senior
Tell me about a video annotation project where your work affected AI model performance. What was your role, and what were the outcomes?
Asked once · last asked Aug 23 · Behavioral · Mid-level
Can you describe a situation where data augmentations might actually reduce the performance of a model?
Asked once · last asked Aug 27 · Scenario · Mid-level
How would you preprocess, clean, and augment a large, noisy, or poorly labeled image dataset to improve model quality and robustness?
Asked once · last asked Aug 27 · Behavioral · Mid-level
Other ways it was asked
- “Tell us about a time you had to work with a large, noisy, or poorly labeled image dataset. What steps did you take to preprocess, clean, and augment the data to ensure the quality and robustness of your model?”
Can you walk me through the end-to-end process of converting raw audio data into time series features?
Asked once · last asked Aug 27 · Concept · Mid-level
What preprocessing steps, such as background-noise filtering, are useful for audio data?
Asked once · last asked Aug 27 · Concept · Junior
How would you process image data into ML-ready features, from raw images through the preprocessing and feature-extraction pipeline?
Asked once · last asked Aug 27 · Concept · Mid-level
Other ways it was asked
- “How did you process the image data—can you walk me through the pipeline of converting image data into ML-ready features?”
When setting up a predictive model for user retention, how do you decide which features are most important to include? Walk me through your typical approach.
Asked once · last asked Aug 29 · Scenario · Mid-level
Why is it critical to align feature logic to business definitions, and how does it impact the overall effectiveness of data models?
Asked once · last asked Aug 31 · Concept · Senior
What steps do you take to prepare collected data for machine learning?
Asked once · last asked Sep 10 · Scenario · Mid-level
How would you encode categorical features such as education level for a machine-learning model?
Asked once · last asked Sep 10 · Write code · Junior
How do you decide which numerical value to assign to each category when encoding categorical data?
Asked once · last asked Sep 10 · Concept · Junior
Can machine-learning models be trained directly on string-valued categorical features?
Asked once · last asked Sep 10 · Concept · Junior
What would be your thought process for creating a dataset for transfer learning?
Asked once · last asked Sep 16 · Scenario · Mid-level
What additional features or signals would you consider to model enterprise client health or return on spend?
Asked once · last asked Sep 18 · Case · Senior
When is normalization useful in linear regression, and how does it make features like age and income comparable?
Asked once · last asked Sep 20 · Concept · Junior
Other ways it was asked
- “Why is normalization important in linear regression, and how does it make features like age and income comparable?”
How do you apply normalization in pandas?
Asked once · last asked Sep 20 · Tooling · Junior
How do you implement binning in Python?
Asked once · last asked Sep 20 · Tooling · Junior
How do you convert a categorical variable like fuel type into a numerical format using one-hot encoding in pandas?
Asked once · last asked Sep 20 · Tooling · Junior
Can a machine-learning model consume collected data directly, or does the data need preprocessing first?
Asked once · last asked Sep 22 · Concept · Junior
How should data be handled between collection and input to a machine-learning model?
Asked once · last asked Sep 22 · Concept · Junior
How would you structure categories for a classification model, including whether to use a catch-all category?
Asked once · last asked Sep 22 · Scenario · Senior
If the dataset is unbalanced, what would you do before training?
Asked once · last asked Sep 23 · Scenario · Mid-level
What would you do if a feature behaves opposite to business expectations?
Asked once · last asked Sep 29 · Scenario · Mid-level
What preprocessing steps are useful before extracting attributes from text and images?
Asked once · last asked Oct 6 · Scenario · Mid-level
What labels would you include in a manually annotated ground-truth dataset for retail product attribute extraction?
Asked once · last asked Oct 6 · Scenario · Mid-level
How do you handle outliers, correlated features, and missing values during feature engineering?
Asked once · last asked Oct 9 · Concept · Mid-level
For a predictive data analysis project, how do you determine which features to engineer and which models to evaluate?
Asked once · last asked Oct 15 · Scenario · Mid-level
How would you determine whether missing occupant or driver fields are data errors or legitimately unknown values, and decide whether to drop, impute, or flag them before modeling?
Asked once · last asked Oct 15 · Scenario · Mid-level
Other ways it was asked
- “How do you determine whether missing fields are data errors or legitimately unknown values, and decide whether to drop, impute, or flag them before modeling?”
If a feature column has many missing values, how would you handle them, and what other processes would you use to assess data quality and prepare data for a model or analysis?
Asked once · last asked Oct 18 · Scenario · Mid-level
What data quality checks would you apply to a data source before delivering a table or machine learning model?
Asked once · last asked Oct 18 · Scenario · Mid-level
If an important feature has many missing values, how would you clean the data and recover useful information?
Asked once · last asked Oct 19 · Scenario · Mid-level
How would you distinguish customer churn based on inactivity from churn based on reduced transaction frequency?
Asked once · last asked Oct 20 · Case · Mid-level
How would you choose a prediction horizon, such as 30 or 90 days, for a churn model?
Asked once · last asked Oct 20 · Case · Mid-level
How would you choose the observation window for a churn prediction model?
Asked once · last asked Oct 20 · Case · Mid-level
For a churn prediction dataset, would you use one row per customer or multiple rows per customer, and why?
Asked once · last asked Oct 20 · Case · Mid-level
How would you group churn prediction data by customer and month?
Asked once · last asked Oct 20 · Case · Mid-level
How do you calculate the positive rate in a churn prediction dataset?
Asked once · last asked Oct 20 · Case · Mid-level
When you’re working with clinical imaging data, such as lung CT scans, how do you ensure the data is well prepared for model training and validation?
Asked once · last asked Oct 20 · Scenario · Mid-level
When building a three-model system with cross-modal features, how would you determine which features are most informative for each model?
Asked once · last asked Oct 20 · Behavioral · Senior
Other ways it was asked
- “When building a three-model system with cross-modal features, how did you determine which features were most informative for each model?”
How would you handle missing dissolved oxygen and pH values in an ocean-sample dataset?
Asked once · last asked Oct 21 · Scenario · Mid-level
What features or signals would you use to determine whether a banking transaction is fraudulent?
Asked once · last asked Oct 22 · Case · Mid-level
Which independent variables would you consider for an initial multiple linear regression model of electricity bills?
Asked once · last asked Oct 24 · Case · Junior
How would you handle severe class imbalance in logistic regression while keeping predicted probabilities well calibrated?
Asked once · last asked Oct 23 · Scenario · Senior
After encoding categorical values with a dictionary, what would you do next to prepare the data for model training?
Asked once · last asked Sep 10 · Scenario · Junior
Why do machine-learning models often require categorical data to be converted into numbers?
Asked once · last asked Sep 10 · Concept · Junior
What do you understand by “features” in machine learning?
Asked once · last asked Mar 13 · Concept · Junior
What impact can the number of features in a dataset have on a machine learning model?
Asked once · last asked Mar 25 · Concept · Junior
Are features added in rows or columns?
Asked once · last asked Mar 25 · Concept · Junior
Is removing missing values from a dataset the right way to handle them?
Asked once · last asked May 25 · Scenario · Junior
What is the purpose of producing a structured labeled dataset?
Asked once · last asked Nov 13 · Concept · Junior
What are the biggest mistakes in data mining?
Asked once · last asked Aug 27 · Concept · Junior
How would you prepare a downloaded CSV dataset for machine-learning training?
Asked once · last asked Sep 10 · Scenario · Junior
What does encoding mean in machine learning?
Asked once · last asked Sep 10 · Concept · Junior
How would you encode a plan tier column as numeric values?
Asked once · last asked Sep 18 · Write code · Junior
What is data preprocessing?
Asked once · last asked Sep 20 · Concept · Junior
What is data normalization?
Asked once · last asked Sep 20 · Concept · Junior
What are three common methods for normalizing numerical features?
Asked once · last asked Sep 20 · Concept · Junior
Other ways it was asked
- “What are the three normalization methods?”
What is binning?
Asked once · last asked Sep 20 · Concept · Junior
What is categorical variable encoding?
Asked once · last asked Sep 20 · Concept · Junior
In one-hot encoding for fuel type, what values would the gas and diesel features have for a car that uses diesel?
Asked once · last asked Sep 20 · Concept · Junior
How should missing data be handled in a machine learning pipeline?
Asked once · last asked Oct 9 · Scenario · Mid-level
What would the dependent variable (Y) represent in an electricity-bill prediction model?
Asked once · last asked Oct 24 · Case · Junior
More Machine Learning interview questions
- All Machine Learning interview questions
- Machine Learning Fundamentals Interview Questions
- Machine Learning Algorithms and Model Selection Interview Questions
- Model Training and Hyperparameter Tuning Interview Questions
- Cross-validation and Data Leakage Interview Questions
- Model Evaluation Metrics Interview Questions
- Applied Machine Learning Interview Questions
- Search and Recommendation System Interview Questions
- MLOps and Model Deployment 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