Selecting Classiers for an Ensemble ◾ 337
EM
j
incorrectly predicts observation “i.” It should be noted that w
rj
>0 indicates
that base classier “r” is included in ensemble EM
j
.
FWA Model
e FWA model involves the averaging of the posterior probabilities of each con-
stituent base classier of the ensemble. Let ρ
ijr
be the posterior probability that base
classier “r” assigns to the actual class of observation “i” of EL
j
. Another approach
to selecting the set of base classiers to include in the EM
j
involves solving the fol-
lowing mixed IP problem:
Min ∑
i∈Ij
y
ij
such that
3.2.3a: ∑
r∈R
(ρ
ijr(*)
– ρ
ijr(c)
)w
rj
+ y
ij
≥ 0 ∀ i ∈ EL
j
, ∀ c ∈ (C – c
(i)
)
3.2.3b: ∑
r∈R
w
rj
= 1
3.2.3c: w
rj
≥ 0 ∀ r ∈ R
3.2.3d: y
ij
∈ {0, 1} ∀ i ∈ EL
j
where ρ
ijr(*)
is the posterior ...