Area under the ROC Curve by Bubble-Sort Approach (BSA)

Area under the ROC Curve by Bubble-Sort Approach (BSA)

Paper in proceedings (conference paper)

A new approach to area under ROC curve (AUC) evaluation is introduced and compared with the current methods. The main idea is based on the Bubble-Sort method. It advantages from the approach to the qualitative dependent variable which is used as the ordinal (and not nominal) variable in comparison to the classical approach. For binary output data the algorithm reaches the same complexity and values as the other methods. For multi-class classification the complexity differs from the classical approaches and the BSA values of AUC don't fail in the special cases as the current methods do.

A new approach to area under ROC curve (AUC) evaluation is introduced and compared with the current methods. The main idea is based on the Bubble-Sort method. It advantages from the approach to the qualitative dependent variable which is used as the ordinal (and not nominal) variable in comparison to the classical approach. For binary output data the algorithm reaches the same complexity and values as the other methods. For multi-class classification the complexity differs from the classical approaches and the BSA values of AUC don't fail in the special cases as the current methods do.

ROC, AUC, AUC evaluation, computational complexity, classification

HONZÍK, P.

13.03.2005

WSEAS

Praha

960-8457-12-2

Automatic Control, Modeling and Simulation (ACMOS'05)

494

6

http://www.wseas.us/e-library/conferences/2005prague/papers/493-299.pdf

http://hdl.handle.net/