Defining optimal cut-off points for multiple class ROC analysis: generalization of the Index of Union method.

Area of Science:

• Machine Learning
• Statistical Modeling
• Data Analysis

Background:

• Binary classification methods are well-established.
• Extensions exist for multi-class settings, but often lack generalizability.
• The Index of Union (IU) method showed prior effectiveness in binary classification.

Purpose of the Study:

• To generalize the Index of Union (IU) method for multi-class classification.
• To evaluate the performance of the Generalized Index of Union (GIU) method.
• To compare GIU against existing multi-class classification techniques.

Main Methods:

• Generalization of the Index of Union (IU) to create the Generalized Index of Union (GIU) method.
• Comparative analysis of GIU with existing methods using simulated datasets.
• Validation of GIU performance on real-world datasets.

Main Results:

• The GIU method demonstrated effectiveness across diverse scenarios and data distributions.
• Performance was robust even with high Volume Under the Surface (VUS) values.
• GIU proved comparable or superior to existing multi-class classification methods.

Conclusions:

• The Generalized Index of Union (GIU) is a versatile and effective method for multi-class classification.
• GIU offers a computationally simple and fast approach for determining optimal cut-off points in ROC analyses.
• The method is recommended for broad application in ROC analysis across all data distributions.