Accuracy and cut-off point selection in three-class classification problems using a generalization of the Youden...

Area of Science:

• Statistics
• Machine Learning
• Biostatistics

Background:

• Assessing classification accuracy is crucial in statistical analysis.
• Existing metrics like the Youden index are limited to two-class problems.
• There is a need for a generalized accuracy index for multi-class scenarios.

Purpose of the Study:

• To define and study the properties of the J(3) index for three-class classification.
• To evaluate the J(3) index as a generalization of the Youden index.
• To apply the J(3) index for assessing discrimination in real-world data.

Main Methods:

• Definition of the J(3) index as maximum correct classification.
• Identification of optimal cut-off points (c(1), c(2)) to maximize correct classification proportions.
• Consideration of both parametric and non-parametric estimation and testing approaches.

Main Results:

• The J(3) index quantifies discrimination between three distributions.
• Optimal cut-off points are determined to maximize overall correct classification.
• The J(3) index effectively generalizes the Youden index for three-class problems.

Conclusions:

• The J(3) index provides a robust measure of accuracy for three-class classification.
• It facilitates optimal threshold selection for enhanced diagnostic performance.
• The index is applicable to medical data, such as studies on human immunodeficiency virus (HIV) patients.