Information Technology Reference
In-Depth Information
ulations; therefore, their confidences are independent and their global confidence
is the product probability of independent confidences, as shown in the following:
n
(
{
t
1
,
t
2
,...,
t
n
}
)
=
(
t
i
).
GlobConf
TermConf
(9.4)
i
=
1
On the contrary, action terms that belong to the same meta-action are extracted
from the same object population, and might be extracted from the same objects.
Therefore, such action terms are dependent on the objects and their global confidence
is dependent on their transition probability. To avoid confusion, and for the sake
of simplicity, we will define the global confidence of action terms from the same
meta-action as follows:
Like
sup
Left
n
n
GlobConf
(
{
t
1
,
t
2
,...,
t
n
}
)
=
1
{
t
i
}
/
1
{
t
i
}
.
(9.5)
i
=
i
=
Global confidence will inform us on how well we can trigger the antecedent side
of an action rule; however, it does not inform us about how well the features values
transitioned by the meta-action will cascade into the desired decision feature value.
For this reason, we define a new metric called execution confidence
ExConf
that
computes the confidence of execution of an action rule. The execution confidence of
an action rule
r
triggered by the set of meta-actions
m
is as follows:
ExConf
(
m
,
r
)
=
GlobConf
(
m
(
r
))
·
Conf
(
r
),
(9.6)
(
)
where
m
represents the set of action terms used in triggering the antecedent side
of the action rule
r
.
With the introduction of the
ExConf
, one could select the action rules with the
highest execution confidence, along with their corresponding meta-actions. Doing
so would insure that the action rules chosen are more likely to be accurate. However,
decision makers may also be interested in the highest return on investment solution,
which would be good enough to solve the issue in hand. In fact, meta-actions are
commonly associated with a cost based on the domain of interest. For example, in
the healthcare domain, each treatment is associated with a cost, and patients are
discharged with a bill including all their medical expenses.
Let us assume that cost
C
i
is associated with each meta-action
m
i
r
)
used. Then a good measure associating the cost
C
i
and the execution confidence
ExConf
would be the satisfaction rate noted as
SatRate
. The satisfaction rate gives a
pointer to which action rule
r
is good enough; in other words, which action rule and
corresponding meta-actions incurs the minimum cost while returning an acceptable
execution confidence. The satisfaction rate for a rule
r
and a set of corresponding
meta-actions
M
r
is computed as follows:
∈
M
(
S
Search WWH ::
Custom Search