CAIC Data Analytics 3 — Questions and Answers
Question 1: Which data analytics concept describes the idea that correlations in data do not establish that one variable causes changes in another?
- Multicollinearity
- Spurious correlation
- Causality vs. correlation (Correct answer)
- Confounding bias
Correct answer: Causality vs. correlation
The causality vs. correlation principle reminds analysts that a statistical relationship does not prove a directional causal mechanism.
Question 2: An AI project team wants to reduce the number of input features before model training. Which technique preserves the most variance while compressing dimensions?
- One-hot encoding
- Principal Component Analysis (PCA) (Correct answer)
- Min-max scaling
- SMOTE
Correct answer: Principal Component Analysis (PCA)
PCA projects data onto orthogonal components ranked by explained variance, compressing dimensions while retaining the most information.
Question 3: What is the primary purpose of a confusion matrix in evaluating a classification model?
- To rank feature importance
- To measure training speed
- To break down correct and incorrect predictions by class (Correct answer)
- To detect data drift
Correct answer: To break down correct and incorrect predictions by class
A confusion matrix tabulates true positives, false positives, true negatives, and false negatives, revealing where and how a classifier makes errors.
Question 4: A client's analytics dashboard shows a sudden spike in daily active users. Before concluding this is real growth, what should an AI consultant check first?
- Model hyperparameters
- Data pipeline logs for tracking anomalies or bugs (Correct answer)
- Feature importance scores
- Gradient descent convergence
Correct answer: Data pipeline logs for tracking anomalies or bugs
Unusual spikes in metrics often result from tracking code bugs, duplicate event firing, or bot traffic rather than genuine user growth.
Question 5: Which sampling strategy should be used when the target class makes up only 1% of the dataset in a binary classification problem?
- Simple random sampling
- Stratified sampling with oversampling of the minority class (Correct answer)
- Systematic sampling
- Cluster sampling
Correct answer: Stratified sampling with oversampling of the minority class
Stratified sampling combined with oversampling (e.g., SMOTE) ensures the minority class is adequately represented, preventing models from ignoring it.
Question 6: What does the term 'data lineage' refer to in an enterprise analytics context?
- The age of a dataset
- The documented history of data's origin, movement, and transformations (Correct answer)
- The number of columns in a table
- The schema version of a database
Correct answer: The documented history of data's origin, movement, and transformations
Data lineage tracks where data came from, how it was transformed at each step, and where it flows, enabling auditability and debugging.
Question 7: An AI consultant wants to evaluate whether two datasets from different time periods have the same statistical distribution. Which test is most appropriate?
- Chi-square goodness-of-fit test
- Kolmogorov-Smirnov test (Correct answer)
- Pearson correlation
- T-test for means
Correct answer: Kolmogorov-Smirnov test
The Kolmogorov-Smirnov test compares the cumulative distribution functions of two samples to detect distributional differences without assuming normality.
CAIC Data Analytics 3
Which data analytics concept describes the idea that correlations in data do not establish that one variable causes changes in another?