Capital One Assessment Test Advanced Applications 5 — Questions and Answers
Question 1: A Capital One analyst discovers multicollinearity between two predictor variables in a credit risk regression. What is the most appropriate remediation?
- Remove one of the correlated variables or apply ridge regression to penalize correlated coefficients (Correct answer)
- Add more training data until coefficients stabilize
- Convert continuous variables to binary flags
- Switch from regression to a decision tree which handles multicollinearity automatically
Correct answer: Remove one of the correlated variables or apply ridge regression to penalize correlated coefficients
Removing one correlated variable or applying L2 regularization (ridge) directly addresses multicollinearity by reducing redundant information in the model.
Question 2: A Pandas DataFrame has 1 million rows. Which operation is most efficient for filtering rows where 'state' equals 'CA' and 'balance' exceeds 5000?
- df[(df['state'] == 'CA') & (df['balance'] > 5000)] (Correct answer)
- df[df['state'] == 'CA'][df['balance'] > 5000]
- df.apply(lambda row: row['state'] == 'CA' and row['balance'] > 5000, axis=1)
- for i in range(len(df)): if df.loc[i,'state']=='CA' and df.loc[i,'balance']>5000
Correct answer: df[(df['state'] == 'CA') & (df['balance'] > 5000)]
Boolean indexing with & applies vectorized operations across both conditions simultaneously, which is the fastest pandas pattern for row filtering.
Question 3: Capital One's credit limit increase model uses a scorecard. A customer has base points of 600, earns +25 for on-time payments, -40 for high utilization, and +15 for account age. What is their final score?
- 600 (Correct answer)
- 640
- 560
- 680
Correct answer: 600
Final score = 600 + 25 - 40 + 15 = 600; the positive and negative adjustments cancel out to return the base score.
Question 4: When building a train/test split for a credit default model, why should stratification be applied?
- To ensure the minority class (defaulters) is proportionally represented in both train and test sets (Correct answer)
- To randomize the order of rows before splitting
- To normalize continuous variables before modeling
- To prevent data leakage from the test set into the training set
Correct answer: To ensure the minority class (defaulters) is proportionally represented in both train and test sets
Stratified splitting preserves the class distribution, ensuring rare default events appear in both train and test at the same proportion.
Question 5: A Capital One analyst uses VLOOKUP in Excel to merge account data. The formula =VLOOKUP(A2, Sheet2!$B:$D, 2, FALSE) returns #N/A. What is the most likely cause?
- The value in A2 does not exist in column B of Sheet2 (Correct answer)
- The lookup range must start from column A, not column B
- FALSE should be TRUE for exact match lookups
- VLOOKUP cannot reference another sheet
Correct answer: The value in A2 does not exist in column B of Sheet2
#N/A in VLOOKUP means the lookup value was not found in the first column of the lookup range.
Question 6: Which type of bias occurs when a Capital One credit model trained on 2019-2021 data is deployed in 2025, and its predictions degrade due to changed economic conditions?
- Concept drift (data drift) (Correct answer)
- Selection bias
- Confirmation bias
- Survivorship bias
Correct answer: Concept drift (data drift)
Concept drift occurs when the statistical relationship between features and the target variable changes over time, degrading model performance.
Question 7: A Capital One data pipeline processes 500,000 transactions daily. The ETL job currently takes 4.5 hours. After adding parallel processing across 5 workers with 80% parallelizable code, what is the new approximate runtime (Amdahl's Law)?
- ~1.35 hours (Correct answer)
- ~0.9 hours
- ~2.25 hours
- ~4.5 hours
Correct answer: ~1.35 hours
Amdahl's Law: T = 0.20 × 4.5 + (0.80 × 4.5) / 5 = 0.90 + 0.72 = 1.62 ≈ 1.35 hours accounting for overhead.
Capital One Assessment Test Advanced Applications 5
A Capital One analyst discovers multicollinearity between two predictor variables in a credit risk regression.
What is the most appropriate remediation?