A logistic regression model is trained on a severely imbalanced dataset where 95% of samples are negative. The model predicts negative for every input. What metric would reveal this failure while accuracy would not?