Obuchowski And Rockette Method

TL;DR

• A technique for comparing two or more diagnostic tests or classifiers by evaluating their ROC curves.
• Uses the area under the ROC curve (AUC) for each test to determine which is more accurate.
• Interprets AUC values where 0.5 indicates no accuracy and 1 indicates perfect accuracy.

Definition

The Obuchowski and Rockette method is a statistical method for evaluating and comparing the accuracy of diagnostic tests or classifiers based on their receiver operating characteristic (ROC) curves by calculating and comparing the area under each ROC curve (AUC).

Explanation

• A ROC curve plots the true positive rate (sensitivity) against the false positive rate for different thresholds of a diagnostic test or classifier.
• The false positive rate is expressed as:

$$\text{False positive rate} = 1 - \text{specificity}$$

• The true positive rate is the proportion of actual positive cases correctly identified; the false positive rate is the proportion of actual negative cases incorrectly identified as positive.
• The AUC summarizes overall accuracy: a value of 0.5 indicates no accuracy and a value of 1 indicates perfect accuracy.
• The Obuchowski and Rockette method compares the AUCs from two or more tests or classifiers to determine which is more accurate.

Examples

Example 1

Suppose you are trying to evaluate the accuracy of two different diagnostic tests for a certain disease. Test A has an AUC of 0.75, while Test B has an AUC of 0.80. According to the Obuchowski and Rockette method, Test B is more accurate than Test A because it has a higher AUC.

Example 2

Suppose you are trying to evaluate the accuracy of two different classifiers for identifying spam emails. Classifier A has an AUC of 0.65, while Classifier B has an AUC of 0.70. According to the Obuchowski and Rockette method, Classifier B is more accurate than Classifier A because it has a higher AUC.

Use cases

• Evaluating the accuracy of diagnostic tests or classifiers by comparing their ROC curves and AUC values.

Related terms

• ROC curve (receiver operating characteristic curve)
• AUC (area under the ROC curve)