Logistic Regression and Classification Metrics Worksheet

Which statement best explains why logistic regression is usually preferred over linear regression for binary classification?

• Logistic regression always gives a smaller training error.

• Logistic regression produces outputs that can be interpreted as probabilities between 0 and 1.

• Logistic regression can only be used with one feature.

• Logistic regression does not require a target variable.

In binary classification, what does a predicted probability of 0.82 usually mean?

• The model is 82% accurate overall.

• There is an 82% chance the observation belongs to the positive class, according to the model.

• 82% of the features are important.

• The model made 82 predictions correctly.

A classifier uses a threshold of 0.5 on predicted probability. What happens if the predicted probability for class 1 is 0.47?

• The model predicts class 1.

• The model predicts class 0.

• The model refuses to classify the case.

• The model changes the threshold automatically.

What is the main role of the logistic function in logistic regression?

• It sorts the training data by class.

• It converts a linear score into a value between 0 and 1.

• It removes outliers from the dataset.

• It standardises the input features.

Which of the following is the best interpretation of training accuracy?

• Performance on unseen data only

• The proportion of correct predictions on the data used to fit the model

• The probability that the model is mathematically correct

• The number of features selected by the model

Why is evaluating a classifier only on the training data potentially misleading?

• Training data is always too noisy to use

• The model may appear better than it really is because it has already seen those examples

• Accuracy cannot be computed on training data

• Training data can only be used for regression

What information does a confusion matrix provide?

• A list of feature coefficients only

• A summary of correct and incorrect predictions by class

• A chart of probability thresholds

• A ranking of the most useful rows in the dataset

Which metric focuses on how many of the predicted positive cases were actually positive?

• Recall

• Precision

• Accuracy

• Mean squared error

What is the main advantage of cross-validation compared with using a single train/test split?

• It guarantees perfect generalisation

• It usually gives a more stable estimate of model performance

• It removes the need for a test set in all situations

• It forces the model to use only one feature

In a binary classification problem, what is an F1-score designed to balance?

• Precision and recall

• Accuracy and runtime

• Mean and standard deviation

• Training size and test size