Evaluation function d: distributional
unusualness
Adapting propositional measures from [Kloesgen/96].
For database D, object relation ro with key
attributes K and binary goal attribute Ag, define
-
c(h) := p[K]({s
| hs Î D}),
the coverage of h
-
T := {t Î ro
| t[Ag] = 1, the target objects
-
g(h) := |c(h)| / |r0|, the generality of h
-
p0 := |T| / |ro|, the reference probability
of the target group
-
p(h) := |c(h) Ç T| / |c(h)|,
the target group probability in c(h)
The evaluation function is
d(h):= g(h) ^a * (p(h) - p0)
(0 if g < s0)
where a balances between generality and unusualness |
e.g. a=0.5 or a=1
