Recall is the proportion of relevant positive instances that a predictive classification system correctly identifies. Use it to ask how many cases of interest the model finds, rather than relying on accuracy alone. Read it alongside precision and other measures that matter for the decision. section 8.6
Bike Shop prediction proposal
A proposed Bike Shop project names recall and precision as performance measures for both customer-churn and demand predictions. The statement of work makes measuring them part of the project hypothesis, before a system is promised for delivery. section 3.5
When testing and selecting a model:
- Decide which performance measures matter in this context; do not let accuracy stand in for an overall assessment. section 8.6
- Consider recall with precision, F1 score, and relevant nonfunctional test results. section 8.6
- Record how the test results and wider considerations informed the selection, including unresolved trade-offs. section 8.6
