<s>
In	O
statistics	O
and	O
machine	O
learning	O
,	O
ensemble	B-Algorithm
methods	I-Algorithm
use	O
multiple	O
learning	O
algorithms	O
to	O
obtain	O
better	O
predictive	O
performance	O
than	O
could	O
be	O
obtained	O
from	O
any	O
of	O
the	O
constituent	O
learning	O
algorithms	O
alone	O
.	O
</s>
<s>
Unlike	O
a	O
statistical	O
ensemble	O
in	O
statistical	O
mechanics	O
,	O
which	O
is	O
usually	O
infinite	O
,	O
a	O
machine	B-Algorithm
learning	I-Algorithm
ensemble	I-Algorithm
consists	O
of	O
only	O
a	O
concrete	O
finite	O
set	O
of	O
alternative	O
models	O
,	O
but	O
typically	O
allows	O
for	O
much	O
more	O
flexible	O
structure	O
to	O
exist	O
among	O
those	O
alternatives	O
.	O
</s>
<s>
Supervised	B-General_Concept
learning	I-General_Concept
algorithms	O
perform	O
the	O
task	O
of	O
searching	O
through	O
a	O
hypothesis	O
space	O
to	O
find	O
a	O
suitable	O
hypothesis	O
that	O
will	O
make	O
good	O
predictions	O
with	O
a	O
particular	O
problem	O
.	O
</s>
<s>
In	O
one	O
sense	O
,	O
ensemble	B-Algorithm
learning	I-Algorithm
may	O
be	O
thought	O
of	O
as	O
a	O
way	O
to	O
compensate	O
for	O
poor	O
learning	O
algorithms	O
by	O
performing	O
a	O
lot	O
of	O
extra	O
computation	O
.	O
</s>
<s>
Fast	O
algorithms	O
such	O
as	O
decision	B-Algorithm
trees	I-Algorithm
are	O
commonly	O
used	O
in	O
ensemble	B-Algorithm
methods	I-Algorithm
(	O
for	O
example	O
,	O
random	B-Algorithm
forests	I-Algorithm
)	O
,	O
although	O
slower	O
algorithms	O
can	O
benefit	O
from	O
ensemble	O
techniques	O
as	O
well	O
.	O
</s>
<s>
By	O
analogy	O
,	O
ensemble	O
techniques	O
have	O
been	O
used	O
also	O
in	O
unsupervised	B-General_Concept
learning	I-General_Concept
scenarios	O
,	O
for	O
example	O
in	O
consensus	B-Algorithm
clustering	I-Algorithm
or	O
in	O
anomaly	B-Algorithm
detection	I-Algorithm
.	O
</s>
<s>
Many	O
ensemble	B-Algorithm
methods	I-Algorithm
,	O
therefore	O
,	O
seek	O
to	O
promote	O
diversity	O
among	O
the	O
models	O
they	O
combine	O
.	O
</s>
<s>
Although	O
perhaps	O
non-intuitive	O
,	O
more	O
random	O
algorithms	O
(	O
like	O
random	O
decision	B-Algorithm
trees	I-Algorithm
)	O
can	O
be	O
used	O
to	O
produce	O
a	O
stronger	O
ensemble	O
than	O
very	O
deliberate	O
algorithms	O
(	O
like	O
entropy-reducing	O
decision	B-Algorithm
trees	I-Algorithm
)	O
.	O
</s>
<s>
A	O
priori	O
determining	O
of	O
ensemble	O
size	O
and	O
the	O
volume	O
and	O
velocity	O
of	O
big	O
data	O
streams	O
make	O
this	O
even	O
more	O
crucial	O
for	O
online	O
ensemble	B-Algorithm
classifiers	I-Algorithm
.	O
</s>
<s>
Ensemble	O
members	O
can	O
also	O
have	O
limits	O
on	O
the	O
features	O
(	O
e.g.	O
,	O
nodes	O
of	O
a	O
decision	B-Algorithm
tree	I-Algorithm
)	O
,	O
to	O
encourage	O
exploring	O
of	O
diverse	O
features	O
.	O
</s>
<s>
It	O
is	O
illustrated	O
below	O
with	O
an	O
ensemble	O
of	O
four	O
decision	B-Algorithm
trees	I-Algorithm
.	O
</s>
<s>
Random	B-Algorithm
forests	I-Algorithm
like	O
the	O
one	O
shown	O
are	O
a	O
common	O
application	O
of	O
bagging	O
.	O
</s>
<s>
Boosting	B-Algorithm
involves	O
training	O
successively	O
models	O
by	O
emphasizing	O
training	O
data	O
mis-classified	O
by	O
previously	O
learned	O
models	O
.	O
</s>
<s>
In	O
some	O
cases	O
,	O
boosting	B-Algorithm
has	O
yielded	O
better	O
accuracy	O
than	O
bagging	O
,	O
but	O
tends	O
to	O
over-fit	O
more	O
.	O
</s>
<s>
The	O
most	O
common	O
implementation	O
of	O
boosting	B-Algorithm
is	O
Adaboost	B-Algorithm
,	O
but	O
some	O
newer	O
algorithms	O
are	O
reported	O
to	O
achieve	O
better	O
results	O
.	O
</s>
<s>
R	B-Language
packages	O
ensembleBMA	O
and	O
BMA	O
use	O
the	O
prior	O
implied	O
by	O
the	O
,	O
(	O
BIC	O
)	O
,	O
following	O
Raftery	O
(	O
1995	O
)	O
.	O
</s>
<s>
R	B-Language
package	O
BAS	O
supports	O
the	O
use	O
of	O
the	O
priors	O
implied	O
by	O
Akaike	O
information	O
criterion	O
(	O
AIC	O
)	O
and	O
other	O
criteria	O
over	O
the	O
alternative	O
models	O
as	O
well	O
as	O
priors	O
over	O
the	O
coefficients	O
.	O
</s>
<s>
The	O
results	O
from	O
BMA	O
can	O
often	O
be	O
approximated	O
by	O
using	O
cross-validation	B-Application
to	O
select	O
the	O
best	O
model	O
from	O
a	O
bucket	O
of	O
models	O
.	O
</s>
<s>
Likewise	O
,	O
the	O
results	O
from	O
BMC	O
may	O
be	O
approximated	O
by	O
using	O
cross-validation	B-Application
to	O
select	O
the	O
best	O
ensemble	O
combination	O
from	O
a	O
random	O
sampling	O
of	O
possible	O
weightings	O
.	O
</s>
<s>
The	O
most	O
common	O
approach	O
used	O
for	O
model-selection	O
is	O
cross-validation	B-Application
selection	O
(	O
sometimes	O
called	O
a	O
"	O
bake-off	O
contest	O
"	O
)	O
.	O
</s>
<s>
Cross-Validation	B-Application
Selection	O
can	O
be	O
summed	O
up	O
as	O
:	O
"	O
try	O
them	O
all	O
with	O
the	O
training	O
set	O
,	O
and	O
pick	O
the	O
one	O
that	O
works	O
best	O
"	O
.	O
</s>
<s>
Gating	O
is	O
a	O
generalization	O
of	O
Cross-Validation	B-Application
Selection	O
.	O
</s>
<s>
Often	O
,	O
a	O
perceptron	B-Algorithm
is	O
used	O
for	O
the	O
gating	O
model	O
.	O
</s>
<s>
It	O
has	O
been	O
successfully	O
used	O
on	O
both	O
supervised	B-General_Concept
learning	I-General_Concept
tasks	O
(	O
regression	O
,	O
classification	O
and	O
distance	O
learning	O
)	O
and	O
unsupervised	B-General_Concept
learning	I-General_Concept
(	O
density	O
estimation	O
)	O
.	O
</s>
<s>
Weighted	B-Algorithm
majority	I-Algorithm
algorithm	I-Algorithm
(	O
machine	O
learning	O
)	O
.	O
</s>
<s>
R	B-Language
:	O
at	O
least	O
three	O
packages	O
offer	O
Bayesian	O
model	O
averaging	O
tools	O
,	O
including	O
the	O
(	O
an	O
acronym	O
for	O
Bayesian	O
Model	O
Selection	O
)	O
package	O
,	O
the	O
(	O
an	O
acronym	O
for	O
Bayesian	O
Adaptive	O
Sampling	O
)	O
package	O
,	O
and	O
the	O
package	O
.	O
</s>
<s>
Python	B-Language
:	O
scikit-learn	B-Application
,	O
a	O
package	O
for	O
machine	O
learning	O
in	O
Python	B-Language
offers	O
packages	O
for	O
ensemble	B-Algorithm
learning	I-Algorithm
including	O
packages	O
for	O
bagging	O
,	O
voting	O
and	O
averaging	O
methods	O
.	O
</s>
<s>
MATLAB	B-Language
:	O
classification	O
ensembles	O
are	O
implemented	O
in	O
Statistics	O
and	O
Machine	O
Learning	O
Toolbox	O
.	O
</s>
<s>
In	O
recent	O
years	O
,	O
due	O
to	O
growing	O
computational	O
power	O
,	O
which	O
allows	O
for	O
training	O
in	O
large	O
ensemble	B-Algorithm
learning	I-Algorithm
in	O
a	O
reasonable	O
time	O
frame	O
,	O
the	O
number	O
of	O
ensemble	B-Algorithm
learning	I-Algorithm
applications	O
has	O
grown	O
increasingly	O
.	O
</s>
<s>
Some	O
of	O
the	O
applications	O
of	O
ensemble	B-Algorithm
classifiers	I-Algorithm
include	O
:	O
</s>
<s>
Land	O
cover	O
mapping	O
is	O
one	O
of	O
the	O
major	O
applications	O
of	O
Earth	O
observation	O
satellite	O
sensors	O
,	O
using	O
remote	B-Application
sensing	I-Application
and	O
geospatial	B-General_Concept
data	I-General_Concept
,	O
to	O
identify	O
the	O
materials	O
and	O
objects	O
which	O
are	O
located	O
on	O
the	O
surface	O
of	O
target	O
areas	O
.	O
</s>
<s>
Some	O
different	O
ensemble	B-Algorithm
learning	I-Algorithm
approaches	O
based	O
on	O
artificial	B-Architecture
neural	I-Architecture
networks	I-Architecture
,	O
kernel	B-Algorithm
principal	I-Algorithm
component	I-Algorithm
analysis	I-Algorithm
(	O
KPCA	O
)	O
,	O
decision	B-Algorithm
trees	I-Algorithm
with	O
boosting	B-Algorithm
,	O
random	B-Algorithm
forest	I-Algorithm
and	O
automatic	O
design	O
of	O
multiple	O
classifier	O
systems	O
,	O
are	O
proposed	O
to	O
efficiently	O
identify	O
land	O
cover	O
objects	O
.	O
</s>
<s>
Change	B-Application
detection	I-Application
is	O
an	O
image	B-General_Concept
analysis	I-General_Concept
problem	O
,	O
consisting	O
of	O
the	O
identification	O
of	O
places	O
where	O
the	O
land	O
cover	O
has	O
changed	O
over	O
time	O
.	O
</s>
<s>
Change	B-Application
detection	I-Application
is	O
widely	O
used	O
in	O
fields	O
such	O
as	O
urban	O
growth	O
,	O
forest	O
and	O
vegetation	O
dynamics	O
,	O
land	O
use	O
and	O
disaster	O
monitoring	O
.	O
</s>
<s>
The	O
earliest	O
applications	O
of	O
ensemble	B-Algorithm
classifiers	I-Algorithm
in	O
change	B-Application
detection	I-Application
are	O
designed	O
with	O
the	O
majority	O
voting	O
,	O
Bayesian	B-General_Concept
average	I-General_Concept
and	O
the	O
maximum	B-General_Concept
posterior	I-General_Concept
probability	I-General_Concept
.	O
</s>
<s>
By	O
combining	O
the	O
output	O
of	O
single	O
classifiers	O
,	O
ensemble	B-Algorithm
classifiers	I-Algorithm
reduce	O
the	O
total	O
error	O
of	O
detecting	O
and	O
discriminating	O
such	O
attacks	O
from	O
legitimate	O
flash	B-General_Concept
crowds	I-General_Concept
.	O
</s>
<s>
Classification	O
of	O
malware	O
codes	O
such	O
as	O
computer	O
viruses	O
,	O
computer	O
worms	O
,	O
trojans	O
,	O
ransomware	O
and	O
spywares	O
with	O
the	O
usage	O
of	O
machine	O
learning	O
techniques	O
,	O
is	O
inspired	O
by	O
the	O
document	B-Algorithm
categorization	I-Algorithm
problem	I-Algorithm
.	O
</s>
<s>
Ensemble	B-Algorithm
learning	I-Algorithm
systems	O
have	O
shown	O
a	O
proper	O
efficacy	O
in	O
this	O
area	O
.	O
</s>
<s>
An	O
intrusion	O
detection	O
system	O
monitors	O
computer	B-Architecture
network	I-Architecture
or	O
computer	O
systems	O
to	O
identify	O
intruder	O
codes	O
like	O
an	O
anomaly	B-Algorithm
detection	I-Algorithm
process	O
.	O
</s>
<s>
Ensemble	B-Algorithm
learning	I-Algorithm
successfully	O
aids	O
such	O
monitoring	O
systems	O
to	O
reduce	O
their	O
total	O
error	O
.	O
</s>
<s>
Face	O
recognition	O
,	O
which	O
recently	O
has	O
become	O
one	O
of	O
the	O
most	O
popular	O
research	O
areas	O
of	O
pattern	O
recognition	O
,	O
copes	O
with	O
identification	O
or	O
verification	O
of	O
a	O
person	O
by	O
their	O
digital	B-Algorithm
images	I-Algorithm
.	O
</s>
<s>
Hierarchical	O
ensembles	O
based	O
on	O
Gabor	O
Fisher	O
classifier	O
and	O
independent	B-Algorithm
component	I-Algorithm
analysis	I-Algorithm
preprocessing	B-General_Concept
techniques	O
are	O
some	O
of	O
the	O
earliest	O
ensembles	O
employed	O
in	O
this	O
field	O
.	O
</s>
<s>
While	O
speech	B-Application
recognition	I-Application
is	O
mainly	O
based	O
on	O
deep	B-Algorithm
learning	I-Algorithm
because	O
most	O
of	O
the	O
industry	O
players	O
in	O
this	O
field	O
like	O
Google	B-Application
,	O
Microsoft	O
and	O
IBM	O
reveal	O
that	O
the	O
core	O
technology	O
of	O
their	O
speech	B-Application
recognition	I-Application
is	O
based	O
on	O
this	O
approach	O
,	O
speech-based	O
emotion	O
recognition	O
can	O
also	O
have	O
a	O
satisfactory	O
performance	O
with	O
ensemble	B-Algorithm
learning	I-Algorithm
.	O
</s>
<s>
Because	O
ensemble	B-Algorithm
learning	I-Algorithm
improves	O
the	O
robustness	O
of	O
the	O
normal	O
behavior	O
modelling	O
,	O
it	O
has	O
been	O
proposed	O
as	O
an	O
efficient	O
technique	O
to	O
detect	O
such	O
fraudulent	O
cases	O
and	O
activities	O
in	O
banking	O
and	O
credit	O
card	O
systems	O
.	O
</s>
<s>
Therefore	O
,	O
different	O
ensemble	B-Algorithm
classifiers	I-Algorithm
are	O
proposed	O
to	O
predict	O
financial	O
crises	O
and	O
financial	O
distress	O
.	O
</s>
<s>
Also	O
,	O
in	O
the	O
trade-based	O
manipulation	O
problem	O
,	O
where	O
traders	O
attempt	O
to	O
manipulate	O
stock	O
prices	O
by	O
buying	O
and	O
selling	O
activities	O
,	O
ensemble	B-Algorithm
classifiers	I-Algorithm
are	O
required	O
to	O
analyze	O
the	O
changes	O
in	O
the	O
stock	O
market	O
data	O
and	O
detect	O
suspicious	O
symptom	O
of	O
stock	O
price	O
manipulation	O
.	O
</s>
<s>
Ensemble	B-Algorithm
classifiers	I-Algorithm
have	O
been	O
successfully	O
applied	O
in	O
neuroscience	O
,	O
proteomics	O
and	O
medical	O
diagnosis	O
like	O
in	O
neuro-cognitive	O
disorder	O
(	O
i.e.	O
</s>
