← Capital One Interview Insights

Capital One·Data Scientist·Technical Phone Screen·Senior

Senior
May 2026

Summary

Capital One data scientist interview that went deep into regression fundamentals, feature engineering, and evaluation design for a flight delay prediction problem. Five fairly brutal sub-questions all tied to one scenario, so if you blanked on any piece of it the whole thing unraveled.

Questions Asked (5)

Q1

A team used OLS regression to predict a binary delay indicator. Why is OLS the wrong choice here, and what model would you use instead? Walk through the link function, assumptions, and how you'd check them.

Data ModelingTechnical Trade-offs
Author's notes

I knew immediately it should be logistic regression but fumbled the rigorous part.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by explaining why OLS is inappropriate for binary outcomes, focusing on violated assumptions and nonsensical predictions. Then introduce logistic regression as the standard alternative, detailing the logit link function, its assumptions, and diagnostic checks. Conclude with practical considerations like interpretation and model evaluation.

Pro tip: Emphasize that while logistic regression is standard, you should also mention alternatives like probit or tree-based models when appropriate, showing awareness of trade-offs. Also, highlight the importance of checking for separation or complete/quasi-complete separation, a common pitfall in logistic regression.

1. Explain why OLS fails for binary outcomes

Discuss how OLS assumes continuous, unbounded outcomes and homoscedastic errors, which are violated with binary data. Predictions can fall outside [0,1], and the linear probability model misrepresents the relationship.

2. Introduce logistic regression and the logit link

Describe logistic regression as a generalized linear model with a logit link function that maps linear predictor to probabilities in (0,1). Explain that it models the log-odds of the event.

3. Outline assumptions of logistic regression

Cover assumptions: linearity of log-odds with continuous predictors, independence of errors, no perfect multicollinearity, and large sample size for reliable estimates. Mention that homoscedasticity is not required.

4. Describe diagnostic checks

Explain how to check assumptions: use Box-Tidwell for linearity, VIF for multicollinearity, residual plots (e.g., deviance residuals), and goodness-of-fit tests like Hosmer-Lemeshow. Also check for influential points and separation.

5. Discuss model evaluation and alternatives

Mention metrics like AUC-ROC, confusion matrix, and calibration. Briefly note alternatives like probit, complementary log-log, or machine learning models if assumptions are violated or for prediction focus.

Key Points to Mention

  • OLS predictions can exceed [0,1] and assume constant variance, which is violated for binary data.
  • Logistic regression uses the logit link: log(p/(1-p)) = β0 + β1X1 + ... + βkXk.
  • Assumptions: linearity of log-odds, independence, no multicollinearity, and no influential outliers.
  • Diagnostics: Box-Tidwell test, VIF, residual plots, Hosmer-Lemeshow test, and checking for separation.
  • Interpretation: coefficients represent change in log-odds; exponentiate for odds ratios.
  • Alternatives: probit, complementary log-log, or tree-based models for prediction.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q2

The dataset encodes day_of_week as integers 1 through 7 and feeds it directly into the model as a numeric feature. What's wrong with that, and how would you fix it? Also, what statistical test would you run to check whether day actually matters?

Data ModelingRoot Cause Analysis
Author's notes

Pretty classic encoding trap but I spent too long explaining it.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

First, explain that treating day_of_week as a numeric feature imposes an artificial ordinal relationship and equal spacing between days, which is incorrect. Then, describe how to fix it using cyclical encoding (sine/cosine) or one-hot encoding, and finally, recommend a statistical test like ANOVA or a chi-square test to check if day matters.

Pro tip: Mention that the choice between cyclical and one-hot encoding depends on whether the day effect is expected to be smooth or discrete, and that cyclical encoding is often preferred for tree-based models to avoid dimensionality issues.

1. Identify the problem

Explain that encoding day_of_week as integers 1-7 incorrectly assumes a linear, ordinal relationship and equal intervals between days, which can mislead the model.

2. Propose encoding solutions

Suggest one-hot encoding for discrete day effects or cyclical encoding (sine and cosine transformations) to preserve the cyclical nature of days.

3. Discuss model implications

Highlight that the choice of encoding depends on the model type; for linear models, one-hot is common, while for tree-based models, cyclical encoding can be more efficient.

4. Recommend statistical test

Propose ANOVA if the target is continuous and day is categorical, or a chi-square test if the target is categorical, to test whether day_of_week has a significant effect.

5. Validate assumptions

Mention checking assumptions of the test (e.g., normality, homogeneity of variance for ANOVA) and considering post-hoc tests if the overall test is significant.

Key Points to Mention

  • Artificial ordinality and equal spacing assumption
  • One-hot encoding for discrete categories
  • Cyclical encoding using sine and cosine transformations
  • ANOVA for continuous target, chi-square for categorical target
  • Impact on model performance and interpretability
  • Potential need for interaction terms or additional features

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q3

The flight_seats column has negative values from an ETL bug, and there are missing values scattered throughout. How do you handle each, and what are the bias/variance implications of different imputation strategies?

Data ModelingTechnical Trade-offs
Author's notes

Negative seat counts are just invalid so I said treat them as missing rather than trying to clip or transform.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

First, diagnose the root cause and distinguish between negative values (invalid) and missing values (absent). Then, propose handling strategies for each, emphasizing that negative values should be corrected or treated as missing, while missing values require careful imputation that balances bias and variance. Finally, discuss the trade-offs of different imputation methods in terms of model performance and interpretability.

Pro tip: Always validate imputation by checking if the imputed values make sense in the business context (e.g., flight seats cannot be negative) and consider creating an indicator variable for missingness to capture potential information in the missing pattern.

1. Diagnose and Understand the Data Issues

Investigate the extent of negative and missing values, their patterns, and potential causes. Determine if negatives are due to data entry errors, ETL bugs, or valid business cases (e.g., refunds).

2. Handle Negative Values

Since flight seats cannot be negative, treat them as invalid. Options: correct if possible (e.g., absolute value if sign error), set to missing, or cap at zero. Choose based on domain knowledge and impact.

3. Choose Imputation Strategy for Missing Values

Evaluate imputation methods (mean, median, mode, regression, KNN, multiple imputation) based on data type, distribution, and missingness mechanism. Consider bias-variance trade-off: simple imputation (e.g., mean) has low variance but high bias; complex methods reduce bias but may increase variance.

4. Assess Bias-Variance Implications

Explain how each method affects model bias and variance. For example, mean imputation shrinks variance and distorts relationships, while multiple imputation preserves variability but adds computational complexity.

5. Validate and Iterate

Test the impact of chosen strategies on model performance using cross-validation. Monitor for unintended consequences and be prepared to adjust.

Key Points to Mention

  • Distinguish between MCAR, MAR, and MNAR missingness mechanisms and how they influence imputation choice.
  • Negative values in flight_seats are likely invalid; consider setting them to missing or zero after verifying with domain experts.
  • Mean/median imputation reduces variance but introduces bias, especially if data is not MCAR.
  • Multiple imputation or model-based methods can reduce bias but may increase variance and complexity.
  • Create a missing indicator variable to preserve information about missingness.
  • Always validate imputation by comparing distributions and model performance before and after.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q4

Define VIF and derive the formula VIF = 1/(1 - R_j^2). If turnaround_time has a VIF of 12, what does that mean practically? Give three remedies for multicollinearity including regularization, and explain how standardizing features affects coefficient interpretation and VIF diagnostics.

Data ModelingTechnical Trade-offs
Author's notes

The derivation part was rough.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by defining VIF and deriving its formula from the R-squared of regressing one predictor on the others. Then interpret the VIF of 12 in practical terms, list three remedies for multicollinearity, and discuss how standardization affects coefficient interpretation and VIF diagnostics.

Pro tip: Emphasize that VIF is a diagnostic, not a fix; and that standardization changes coefficient units but not VIF values, so it doesn't solve multicollinearity but can help with interpretation and numerical stability.

1. Define VIF and derive formula

Explain that VIF measures how much the variance of an estimated regression coefficient is inflated due to multicollinearity. Derive VIF = 1/(1 - R_j^2) by considering the variance of the j-th coefficient in a multiple regression.

2. Interpret VIF = 12

State that a VIF of 12 means the variance of the coefficient for turnaround_time is 12 times larger than if it were uncorrelated with other predictors. This indicates high multicollinearity, leading to unstable estimates and wide confidence intervals.

3. List remedies for multicollinearity

Provide three remedies: (1) remove or combine correlated predictors, (2) use regularization techniques like ridge regression or LASSO, (3) apply dimensionality reduction such as PCA.

4. Explain effect of standardization

Discuss that standardizing features (e.g., z-score) changes the scale of coefficients, making them comparable in terms of standard deviation units, but does not change VIF values because VIF is based on correlations, which are scale-invariant.

Key Points to Mention

  • VIF formula derivation: VIF_j = 1/(1 - R_j^2), where R_j^2 is from regressing X_j on all other predictors.
  • Practical interpretation: VIF > 10 often indicates problematic multicollinearity; VIF=12 means coefficient variance is inflated 12-fold.
  • Remedies: remove/combine variables, use ridge/LASSO, or PCA.
  • Standardization: coefficients become comparable but VIF remains unchanged; it helps with numerical stability and interpretation.
  • Regularization: ridge regression adds L2 penalty, shrinking coefficients and mitigating multicollinearity effects.
  • Trade-offs: removing variables may cause omitted variable bias; regularization introduces bias but reduces variance.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q5

What evaluation metrics make sense for a binary imbalanced target? Describe a cross-validation strategy that avoids temporal leakage, and explain how you'd calibrate the predicted probabilities.

Product Analytics & MetricsA/B Testing & Experimentation
Author's notes

Went with PR-AUC over ROC-AUC given the imbalance, explained why ROC can look deceptively good with rare positives.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by acknowledging the class imbalance and its impact on metric choice, then recommend metrics like PR-AUC, F1, and recall at a fixed precision, explaining why accuracy is misleading. For cross-validation, emphasize time-series-aware splitting (e.g., expanding window or purged K-fold with embargo) to prevent temporal leakage. Finally, discuss probability calibration methods like Platt scaling or isotonic regression, and how to validate them with reliability diagrams and Brier score.

Pro tip: Tie your metric choice to the business cost of false positives vs. false negatives—Capital One cares about financial impact, so mention how you'd align metrics with cost-sensitive decisions. Also, note that calibration should be done on a separate validation set to avoid overfitting.

1. Choose appropriate evaluation metrics

Explain why accuracy is misleading for imbalanced data and recommend metrics like PR-AUC, F1, recall at high precision, and Matthews correlation coefficient. Discuss how to select based on business objectives.

2. Implement time-aware cross-validation

Describe using expanding window or purged K-fold with embargo to respect temporal order. Highlight that standard K-fold leaks future information and inflates performance.

3. Calibrate predicted probabilities

Discuss methods like Platt scaling (sigmoid) and isotonic regression, and the importance of using a separate calibration set. Mention that tree-based models often need calibration.

4. Validate calibration and metrics

Explain how to assess calibration with reliability diagrams, Brier score, and log loss. Emphasize that calibration should be evaluated on out-of-time data to ensure robustness.

Key Points to Mention

  • Why accuracy, ROC-AUC can be misleading for imbalanced data; PR-AUC is more informative.
  • Business-driven metric selection: cost of false positives vs. false negatives.
  • Time-series cross-validation techniques: expanding window, purged K-fold with embargo.
  • Probability calibration methods: Platt scaling, isotonic regression, and their assumptions.
  • Evaluation of calibration: reliability diagrams, Brier score, log loss.
  • Avoiding data leakage: ensuring calibration and validation sets are temporally separated.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.