NEW
Font size
WorksheetsClassification & Regression - 2
Total questions: 40
Worksheet time: 20mins
Which metric is most important for spam filtering?
Precision
Recall
RMSE
Accuracy
Which metric is most important for disease screening?
Accuracy
Recall
Precision
Specificity
In model evaluation, when both false positives and false negatives are equally costly, the best metric is:
Accuracy
Precision
Recall
F1-score
F1-score is low if:
Both precision and recall are high
Either precision or recall is very low
Accuracy is perfect
Dataset is balanced
In cases where precision = 1 and recall = 0, the F1-score is:
1
0
0.5
Undefined
F1-score is a type of:
Arithmetic mean
Geometric mean
Harmonic mean
Weighted average
Recall is also known as:
Sensitivity
Specificity
Precision
Accuracy
In a medical test, high recall means:
Most healthy people are correctly identified
Most sick people are correctly identified
Few false positives occur
Accuracy is 100%
Recall can be improved at the cost of:
Reducing accuracy
Reducing precision
Reducing F1-score
Increasing false negatives
Precision focuses on reducing:
False negatives
False positives
True negatives
True positives
In a fraud detection system, high precision means:
Most flagged frauds are actually frauds
All frauds are detected
No fraud is missed
System has high accuracy
Precision is also known as:
Sensitivity
Specificity
Positive Predictive Value (PPV)
Negative Predictive Value (NPV)
Accuracy can be misleading when:
Classes are balanced
Classes are imbalanced
Training dataset is large
Model is regularized
In a dataset with 99% negatives and 1% positives, a model predicting all negatives will have accuracy:
100%
99%
50%
1%
Which evaluation metric is better than accuracy for imbalanced datasets?
Precision and Recall
RMSE
R-squared
Variance Inflation Factor
In KNN, smaller values of k usually lead to:
High bias and low variance
Low bias and high variance
Perfect generalization
Always underfitting
Feature scaling (normalization/standardization) is important for KNN because:
Distance calculation depends on scale of features
It prevents overfitting
It improves accuracy always
It reduces training time
Weighted KNN differs from normal KNN by:
Giving equal weight to all neighbors
Giving more weight to closer neighbors
Ignoring distant neighbors completely
Using logistic regression as backend
The main drawback of KNN is:
It requires complex training phase
It is computationally expensive at prediction time
It cannot handle numerical data
It works only for binary classification
The output of logistic regression before applying the sigmoid function is called:
Odds ratio
Logit
Probability
Likelihood
Which of the following is NOT an advantage of logistic regression?
Easy to implement
Works well with linearly separable data
Handles categorical variables
Automatically captures complex nonlinear patterns
Logistic regression can be extended to multi-class classification using:
Linear regression
Softmax regression
Ridge regression
Lasso regression
Logistic regression assumes that:
Relationship between independent and dependent variables is linear in log-odds
Variables are always categorical
The dependent variable is continuous
No correlation exists in data
Which of these is a limitation of classification models?
They cannot predict probabilities
They may be biased by imbalanced datasets
They do not handle categorical features
They are always linear
In classification, the confusion matrix helps in:
Calculating regression coefficients
Summarizing correct & incorrect predictions
Estimating correlation
Removing multicollinearity
Which of the following metrics is scale-independent?
R-squared
MSE
RMSE
MAE
A model with high R² but poor prediction on new data suffers from:
Multicollinearity
Overfitting
Underfitting
Proper fitting
Cross-validation is mainly used to:
Estimate model’s generalization performance
Remove multicollinearity
Reduce correlation
Increase R² artificially
Mean Absolute Error (MAE) is:
Average of squared errors
Average of absolute errors
Root of squared errors
Ratio of errors
Mean Squared Error (MSE) is preferred because:
Easier interpretation
Penalizes larger errors more
Works only with classification
Ignores variance
Adjusted R-squared increases only if:
A new predictor improves the model
A new predictor is always added
A new predictor decreases fit
Sample size reduces
In hypothesis h(x) = β₀ + β₁x, β₁ measures:
Intercept
Average of y
Effect of x on y
Error variance
Residuals in regression represent:
Predicted values
Difference between actual and predicted values
Model coefficients
Independent variables
A fitted regression line is also called:
Hypothesis function
Decision boundary
Error function
Correlation line
One method to handle multicollinearity is:
Adding more correlated variables
Removing highly correlated variables
Ignoring the problem
Increasing p-value threshold
When multicollinearity is high, regression coefficients:
Are very stable
May become unreliable and unstable
Always decrease
Are unaffected
Which of these problems is addressed using regression?
Predicting sales revenue
Classifying spam emails
Grouping customers into clusters
Reducing dimensionality
What is a common issue when using accuracy as a metric for imbalanced datasets?
It requires large datasets
It can be misleading
It is always reliable
It ignores false positives
Which metric is primarily used to evaluate the performance of a model in a binary classification problem?
Mean Squared Error
Adjusted R-squared
F1-score
R-squared
In the context of regression analysis, what does a high R-squared value indicate?
The model is overfitting
The model is linear
The model explains a large proportion of variance
The model has high bias
