Accuracy Can Hide

Positive Recall Is Zero

Use actual positive rows to see the missed P.

highlighted = computed this step

Use actual positives for recall

Positive recall uses TP over TP plus FN. Here that is 0 over 1.

positive recall=01\text{positive recall}={0\over1}
Class Imbalance ExactlyExact row counts for one always-N baseline.label countslabelcountN4P1always-N baseline on one imbalanced tablerowtruepredcorrect?cellANNyesTNBNNyesTNCNNyesTNDNNyesTNEPNmissFNconfusion counts with P as positivecountvalueTP0FN1FP0TN4metric fractionsmetricfractionaccuracy4/5positive recall0/1 = 0accuracy can be high while the lone P is missedone pinned imbalanced table; one always-N baselinethe lone P is missed; this is exact row countingdoes not claim: NOT fairness solved; NOT future performanceNOT calibrated; NOT safe; NOT best metric; NOT accuracy guaranteeNOT generalization; NOT probability truth

The positive recall is zero

The numerator is 0, so the positive recall is 0.

0/1=00/1=0
Class Imbalance ExactlyExact row counts for one always-N baseline.label countslabelcountN4P1always-N baseline on one imbalanced tablerowtruepredcorrect?cellANNyesTNBNNyesTNCNNyesTNDNNyesTNEPNmissFNconfusion counts with P as positivecountvalueTP0FN1FP0TN4metric fractionsmetricfractionaccuracy4/5positive recall0/1 = 0accuracy can be high while the lone P is missedone pinned imbalanced table; one always-N baselinethe lone P is missed; this is exact row countingdoes not claim: NOT fairness solved; NOT future performanceNOT calibrated; NOT safe; NOT best metric; NOT accuracy guaranteeNOT generalization; NOT probability truth

Compare the two summaries

The same table has correct-row fraction 4 over 5 and positive recall 0.

accuracy=4/5,recallP=0\text{accuracy}=4/5,\quad \text{recall}_P=0
Class Imbalance ExactlyExact row counts for one always-N baseline.label countslabelcountN4P1always-N baseline on one imbalanced tablerowtruepredcorrect?cellANNyesTNBNNyesTNCNNyesTNDNNyesTNEPNmissFNconfusion counts with P as positivecountvalueTP0FN1FP0TN4metric fractionsmetricfractionaccuracy4/5positive recall0/1 = 0accuracy can be high while the lone P is missedone pinned imbalanced table; one always-N baselinethe lone P is missed; this is exact row countingdoes not claim: NOT fairness solved; NOT future performanceNOT calibrated; NOT safe; NOT best metric; NOT accuracy guaranteeNOT generalization; NOT probability truth