Confusion Matrix

In a binary classification task, an ML model predicts whether an email is spam or not spam. The positive class is spam and the negative class is not spam. What does a false positive mean?

• The model predicted spam, and the email was actually spam

• The model predicted not spam, and the email was actually not spam

• The model predicted spam, but the email was actually not spam

• The model predicted not spam, but the email was actually spam

A binary classifier predicts whether a patient has a disease. The positive class is has disease. What does precision tell us?

• How many actual disease cases were found by the model

• How trustworthy the model’s “has disease” predictions are

• How many healthy patients were correctly classified

• How many total patients were classified correctly

A binary classifier predicts whether a patient has a disease. The positive class is has disease. What does recall tell us?

• How many predicted disease cases were correct

• How many actual disease cases the model successfully found

• How many healthy patients were incorrectly classified

• How many total patients were in the dataset

A disease detection dataset contains 98% patients who do not have the disease and only 2% who do have the disease. Why can accuracy be misleading in this situation?

• Accuracy only works with image datasets

• A model can get high accuracy by mostly predicting the majority class

• Accuracy always ignores true positives

• Accuracy is only useful for multi-class classification

A classifier predicts whether patients have a disease. The positive class is has disease. If the classifier predicts has disease for every patient, what happens to recall for the disease class?

• Recall becomes 0%

• Recall becomes 50%

• Recall becomes 100%

• Recall cannot be calculated

A classifier predicts whether patients have a disease. The positive class is has disease. If the classifier predicts does not have disease for every patient, what happens to recall for the disease class?

• Recall becomes 100%

• Recall becomes 0%

• Precision becomes 100%

• Accuracy must become 0%

Two models are used to detect faulty products in a factory. The positive class is faulty product. Model A has high precision and high recall. Model B has very high precision and very low recall. Which statement is most accurate?

• Model B is clearly better because precision is higher

• Model A is more balanced because it has both good precision and good recall

• Model B is better because recall does not matter

• Model A is worse because its precision is lower

What is the main purpose of the F1 score in a classification task?

• To measure only the number of true negatives

• To combine precision and recall into one balanced metric

• To replace the confusion matrix entirely

• To measure only the number of correct predictions

In a medical diagnosis system, missing a real disease case is usually serious. In this situation, which metric is especially important?

• Recall

• Accuracy

• Number of true negatives only

• Dataset size

A computer vision model detects product defects. The possible defect classes are scratch, dent, pit mark, and discolouration. In this type of multi-class classification task, how are precision, recall, and F1 score often reported?

• Only once for the whole dataset

• Separately for each class

• Only for the largest class

• Only for the class with the lowest accuracy