ROC Curves for Continuous Data

i

i

“k10031final” — 2009/4/6 — 12:30 — page 6 — #14

i

i

i

i

i

i

6 CHAPTER 1. INTRODUCTION

any object is of the standard class. Again a threshold is adopted, so

that any new object yielding a score which is very unusual is classified

as “nonstandard.”

Note that in all of this the true classes may exist but simply not be

observable at the time of applying the rule, or they may not actually

exist. An example of the former is a medical condition whose presence

cannot be definitively established without a dangerous invasive pro-

cedure. The alternative may be a classification rule based on simple

and quick tests. Eventually the true class does become known (per-

haps post mortem), and a set of such patients can form the training

set. An example of the latter, where