Unit 2: Data Science Methodology

Direct answer: A is correct. Recall asks, “Of all the cases that are truly positive, how many did the model find?” It is calculated as TP ÷ (TP + FN). Here, TP means true positives, or positive cases correctly identified, and FN means false negatives, or positive cases missed by the model. A high recall means that the model misses relatively few actual positive cases. Option B describes precision: among the cases predicted positive, it asks how many are truly positive, using TP ÷ (TP + FP). Option C describes accuracy, which considers all correct predictions among all cases. Option D describes specificity, which measures how well actual negative cases are identified. For a disease-screening example, recall is especially concerned with finding people who really have the disease. Memory cue: recall recalls the real positives; its denominator contains all actual positives, TP plus FN.

Direct answer: A is correct. Overfitting occurs when a model learns the training examples, including accidental noise or unimportant details, so closely that it performs poorly on new data. The model may show very good training performance but weak generalisation. Regularization reduces this problem by adding a penalty for excessive complexity, encouraging the model to use simpler patterns rather than memorising the training set. Depending on the method, the penalty can discourage overly large weights or unnecessary model complexity. Option B is wrong because checking only training data can hide overfitting; a separate validation or test set is needed to examine performance on unseen data. Option C is wrong because duplicate examples do not provide genuinely new information and may reinforce memorisation. Option D is wrong because an untested model cannot be reliably judged. A useful cue is: “Regularization restrains the model.” Cross-validation, simpler models, and suitable data practices can also help, but regularization is the direct choice here.