The ratio of correctly predicted positive observations to the all observations in actual class. It answers: 'Of all the actual positive instances, how many did the model find?'
Critical for medical diagnosis (avoiding missed cases).
Used when the cost of missing a positive is high.