Question

You're training a model to classify whether or not a bridge is likely to collapse given several factors. You have a dataset of thousands of existing bridges and their attributes, where each bridge is labeled as having collapsed or not collapsed. Only a handful of bridges in the dataset are labeled as having collapsed—the rest are labeled as not collapsed. Given your goal of minimizing bridge collapse and the severe harm it can cause, which of the following metrics would be most useful for evaluating the model?

A. Accuracy
B. Recall
C. Precision
D. Confusion matrix

Answer 1

The most useful metric for evaluating a model that classifies whether a bridge is likely to collapse, given the importance of minimizing false negatives due to severe harm, is Recall. This is because Recall measures the proportion of real collapses that are correctly identified, which is critical in this high-stakes context.

Step-by-step explanation:

In the context of classifying whether a bridge is likely to collapse, the most useful metric for evaluating the model is B. Recall. Recall focuses on the model’s ability to correctly identify all actual positives.

While accuracy might seem important, it could be misleading in this case due to the imbalanced nature of the dataset where instances of non-collapsed bridges are much more common than collapsed ones. Precision measures how many of the bridges the model predicts to collapse do actually collapse, but this is less critical than minimizing the number of missed collapses.