DAC Machine Learning & Predictive Analytics 3 — Questions and Answers
Question 1: What does the bias-variance tradeoff describe?
- The balance between training time and accuracy
- The balance between underfitting and overfitting errors (Correct answer)
- The tradeoff between precision and recall
- The choice between supervised and unsupervised learning
Correct answer: The balance between underfitting and overfitting errors
High bias causes underfitting while high variance causes overfitting, and models must balance the two.
Question 2: Which validation approach is most reliable for a small dataset?
- A single 50/50 train-test split
- K-fold cross-validation (Correct answer)
- Using all data for training only
- Testing on the training data
Correct answer: K-fold cross-validation
K-fold cross-validation uses all data for both training and validation across folds, maximizing reliability on small samples.
Question 3: A predictive model performs well in testing but degrades over months in production. This is most likely due to what?
- Data drift (Correct answer)
- Overfitting
- Underfitting
- Feature scaling errors
Correct answer: Data drift
Data drift occurs when the statistical properties of incoming data change over time, degrading model performance.
Question 4: In logistic regression, what does the output represent?
- A continuous numeric prediction
- A probability between 0 and 1 (Correct answer)
- A cluster assignment
- A residual value
Correct answer: A probability between 0 and 1
Logistic regression outputs a probability that an observation belongs to the positive class.
Question 5: Which technique addresses severe class imbalance in a training dataset?
- Increasing the learning rate
- SMOTE oversampling of the minority class (Correct answer)
- Removing the target variable
- Using a larger test set
Correct answer: SMOTE oversampling of the minority class
SMOTE synthesizes new minority-class examples to balance the classes for training.
Question 6: What is the main advantage of a random forest over a single decision tree?
- It is always faster to train
- It reduces variance by averaging many trees (Correct answer)
- It requires no data preprocessing
- It cannot overfit
Correct answer: It reduces variance by averaging many trees
Averaging predictions across many de-correlated trees reduces variance and improves generalization.
Question 7: Which metric summarizes model performance across all classification thresholds?
- Accuracy
- AUC-ROC (Correct answer)
- Mean squared error
- R-squared
Correct answer: AUC-ROC
The AUC-ROC measures a classifier's ability to rank positives above negatives across every threshold.
DAC Machine Learning & Predictive Analytics 3
What does the bias-variance tradeoff describe?