Database Reference
In-Depth Information
Fig. 13.2 Pseudo-code of combining procedure.
(2) “Take-It-All” — In this weighting method, all subsets obtain a weight
that is greater than 0.5. This leads to the situation in which any
attribute that has been in at least one of the subsets will be included
in the final result. This method should have a low false negative rate,
because losing a relevant attribute will take place only if all feature
selections methods will decide to filter out this attribute.
(3) “Smaller is Heavier” — The weight for each selector is defined by
its bias to the smallest subset. Selectors that tend to provide small
subsets will gain more weight than selectors that tend to provide large
subsets. This approach is inspired by the fact that the precision rate
of selectors tend to decrease as the size of the subset increases. This
approach can be used to avoid noise caused by feature selectors that
tend to select most of the possible attributes. More specifically, the
weights are defined as:
ω
|B i |
ω
j =1 |B j |
k =1
w i =
.
(13.11)
|B k |
ω
j =1 |B j |
Search WWH ::




Custom Search