ROC Curves for Continuous Data

i

i

“k10031final” — 2009/4/6 — 12:30 — page 156 — #164

i

i

i

i

i

i

156 CHAPTER 8. BEYOND THE BASICS

Goadrich (2006), Briggs and Zaretski (2008), and others. There seems

to be something about ROC analysis that leads to the occasional re-

invention of the methods: we speculate that it is partly because of

the apparent simplicity of the basic problem, with very few degrees

of freedom having to be represented, and partly because the need for

such representations are so ubiquitous, occurring in widely different

intellectual communities which do not (and cannot be expected to)

read each other’s literature.

The convex hull idea for combining classifiers was described by

Provost and Fawcett (1997, 1998), and Scott et al. (1998), although

there are also traces of it in ear ...