Evaluating biological models
A confusion matrix counts true positives, false positives, true negatives and false negatives.
Loading video…
A confusion matrix counts true positives, false positives, true negatives and false negatives. Sensitivity or recall measures detected positives; specificity measures rejected negatives; precision measures how many positive predictions were truly positive. Accuracy can hide failure on a rare class. ROC/AUC summarizes discrimination across thresholds, but calibration and real costs of errors also matter.
Worked example
For 100 samples with five cases, predicting everyone healthy gives 95% accuracy and 0% sensitivity. A screening task may favor sensitivity while tracking false alarms.