informedness

Informedness is a statistical measure used to assess the quality of a binary classifier or diagnostic test. It captures how well the test’s positive and negative classifications correspond to actual status, accounting for chance. The standard definition is informedness = sensitivity + specificity − 1, where sensitivity is TP/(TP+FN) and specificity is TN/(TN+FP) for a confusion matrix with true positives, false positives, true negatives, and false negatives.

The value of informedness ranges from −1 to +1. A value of +1 indicates perfect information (all positives and negatives classified correctly). A value of 0 indicates information equivalent to random guessing. A value of −1 indicates perfect misclassification (predictions are consistently wrong). Like sensitivity and specificity, informedness depends on the classifier’s performance at a given threshold and is not affected by the prevalence of the positive class.

Origins and relationship to other measures: Informedness is commonly identified with Youden’s J statistic, named after W. J. Youden. It is part of a broader framework introduced by Brennan and Prediger that also includes markedness, with the goal of evaluating test quality that is less influenced by base rates and bias. Informedness complements other metrics such as accuracy, precision, recall, and the ROC curve, providing a threshold-dependent summary of discriminative ability.

Applications and limitations: Informedness is widely used in diagnostic test evaluation, psychometrics, and machine learning to gauge a test’s usefulness beyond chance. It should be interpreted alongside other measures, as it does not capture calibration or cost considerations and can be threshold-dependent.