<s>
An	O
incremental	B-Algorithm
decision	I-Algorithm
tree	I-Algorithm
algorithm	O
is	O
an	O
online	B-Algorithm
machine	I-Algorithm
learning	I-Algorithm
algorithm	O
that	O
outputs	O
a	O
decision	B-Algorithm
tree	I-Algorithm
.	O
</s>
<s>
Many	O
decision	B-Algorithm
tree	I-Algorithm
methods	O
,	O
such	O
as	O
C4.5	B-Algorithm
,	O
construct	O
a	O
tree	O
using	O
a	O
complete	O
dataset	O
.	O
</s>
<s>
Incremental	B-Algorithm
decision	I-Algorithm
tree	I-Algorithm
methods	O
allow	O
an	O
existing	O
tree	O
to	O
be	O
updated	O
using	O
only	O
new	O
individual	O
data	O
instances	O
,	O
without	O
having	O
to	O
re-process	O
past	O
instances	O
.	O
</s>
<s>
Here	O
is	O
a	O
short	O
list	O
of	O
incremental	B-Algorithm
decision	I-Algorithm
tree	I-Algorithm
methods	O
,	O
organized	O
by	O
their	O
(	O
usually	O
non-incremental	O
)	O
parent	O
algorithms	O
.	O
</s>
<s>
CART	O
(	O
1984	O
)	O
is	O
a	O
nonincremental	O
decision	B-Algorithm
tree	I-Algorithm
inducer	O
for	O
both	O
classification	O
and	O
regression	O
problems	O
.	O
</s>
<s>
ID3	B-Algorithm
(	O
1986	O
)	O
and	O
C4.5	B-Algorithm
(	O
1993	O
)	O
were	O
developed	O
by	O
Quinlan	O
and	O
have	O
roots	O
in	O
Hunt	O
's	O
Concept	O
Learning	O
System	O
(	O
CLS	O
,	O
1966	O
)	O
The	O
ID3	B-Algorithm
family	O
of	O
tree	O
inducers	O
was	O
developed	O
in	O
the	O
engineering	O
and	O
computer	O
science	O
communities	O
.	O
</s>
<s>
ID3	B-Algorithm
 '	O
(	O
1986	O
)	O
was	O
suggested	O
by	O
Schlimmer	O
and	O
Fisher	O
.	O
</s>
<s>
It	O
was	O
a	O
brute-force	O
method	O
to	O
make	O
ID3	B-Algorithm
incremental	O
;	O
after	O
each	O
new	O
data	O
instance	O
is	O
acquired	O
,	O
an	O
entirely	O
new	O
tree	O
is	O
induced	O
using	O
ID3	B-Algorithm
.	O
</s>
<s>
ID5	O
(	O
1988	O
)	O
did	O
n't	O
discard	O
subtrees	O
,	O
but	O
also	O
did	O
not	O
guarantee	O
that	O
it	O
would	O
produce	O
the	O
same	O
tree	O
as	O
ID3	B-Algorithm
.	O
</s>
<s>
ID5R	O
(	O
1989	O
)	O
output	O
the	O
same	O
tree	O
as	O
ID3	B-Algorithm
for	O
a	O
dataset	O
regardless	O
of	O
the	O
incremental	O
training	O
order	O
.	O
</s>
<s>
It	O
did	O
not	O
handle	O
numeric	O
variables	O
,	O
multiclass	B-General_Concept
classification	I-General_Concept
tasks	O
,	O
or	O
missing	O
values	O
.	O
</s>
<s>
ID6MDL	O
(	O
2007	O
)	O
an	O
extended	O
version	O
of	O
the	O
ID3	B-Algorithm
or	O
ID5R	O
algorithms	O
.	O
</s>
<s>
There	O
were	O
several	O
incremental	O
concept	O
learning	O
systems	O
that	O
did	O
not	O
build	O
decision	O
trees	O
,	O
but	O
which	O
predated	O
and	O
influenced	O
the	O
development	O
of	O
the	O
earliest	O
incremental	B-Algorithm
decision	I-Algorithm
tree	I-Algorithm
learners	O
,	O
notably	O
ID4	O
.	O
</s>
<s>
STAGGER	O
was	O
developed	O
to	O
examine	O
concepts	O
that	O
changed	O
over	O
time	O
(	O
concept	B-Algorithm
drift	I-Algorithm
)	O
.	O
</s>
<s>
Prior	O
to	O
STAGGER	O
,	O
Michalski	O
and	O
Larson	O
(	O
1978	O
)	O
investigated	O
an	O
incremental	O
variant	O
of	O
AQ	O
(	O
Michalski	O
,	O
1973	O
)	O
,	O
a	O
supervised	O
system	O
for	O
learning	O
concepts	O
in	O
disjunctive	B-Application
normal	I-Application
form	I-Application
(	O
DNF	O
)	O
.	O
</s>
<s>
Experience	O
with	O
these	O
earlier	O
systems	O
and	O
others	O
,	O
to	O
include	O
incremental	O
tree-structured	O
unsupervised	O
learning	O
,	O
contributed	O
to	O
a	O
conceptual	O
framework	O
for	O
evaluating	O
incremental	B-Algorithm
decision	I-Algorithm
tree	I-Algorithm
learners	O
specifically	O
,	O
and	O
incremental	O
concept	O
learning	O
generally	O
,	O
along	O
four	O
dimensions	O
that	O
reflect	O
the	O
inherent	O
tradeoffs	O
between	O
learning	O
cost	O
and	O
quality	O
:	O
(	O
1	O
)	O
cost	O
of	O
knowledge	O
base	O
update	O
,	O
(	O
2	O
)	O
the	O
number	O
of	O
observations	O
that	O
are	O
required	O
to	O
converge	O
on	O
a	O
knowledge	O
base	O
with	O
given	O
characteristics	O
,	O
(	O
3	O
)	O
the	O
total	O
effort	O
(	O
as	O
a	O
function	O
of	O
the	O
first	O
two	O
dimensions	O
)	O
that	O
a	O
system	O
exerts	O
,	O
and	O
the	O
(	O
4	O
)	O
quality	O
(	O
often	O
consistency	O
)	O
of	O
the	O
final	O
knowledge	O
base	O
.	O
</s>
<s>
Some	O
of	O
the	O
historical	O
context	O
in	O
which	O
incremental	B-Algorithm
decision	I-Algorithm
tree	I-Algorithm
learners	O
emerged	O
is	O
given	O
in	O
Fisher	O
and	O
Schlimmer	O
(	O
1988	O
)	O
,	O
and	O
which	O
also	O
expands	O
on	O
the	O
four	O
factor	O
framework	O
that	O
was	O
used	O
to	O
evaluate	O
and	O
design	O
incremental	B-Algorithm
learning	I-Algorithm
systems	O
.	O
</s>
<s>
Very	O
Fast	O
Decision	O
Trees	O
learner	O
reduces	O
training	O
time	O
for	O
large	O
incremental	O
data	O
sets	O
by	O
subsampling	O
the	O
incoming	O
data	B-General_Concept
stream	I-General_Concept
.	O
</s>
<s>
CVFDT	O
(	O
2001	O
)	O
can	O
adapt	O
to	O
concept	B-Algorithm
drift	I-Algorithm
,	O
by	O
using	O
a	O
sliding	O
window	O
on	O
incoming	O
data	O
.	O
</s>
<s>
VFDTc	O
(	O
2006	O
)	O
extends	O
VFDT	O
for	O
continuous	O
data	O
,	O
concept	B-Algorithm
drift	I-Algorithm
,	O
and	O
application	O
of	O
Naive	O
Bayes	O
classifiers	O
in	O
the	O
leaves	O
.	O
</s>
<s>
The	O
Extremely	O
Fast	O
Decision	B-Algorithm
Tree	I-Algorithm
learner	O
is	O
statistically	O
more	O
powerful	O
than	O
VFDT	O
,	O
allowing	O
it	O
to	O
learn	O
more	O
detailed	O
trees	O
from	O
less	O
data	O
.	O
</s>
<s>
During	O
incremental	B-Algorithm
learning	I-Algorithm
this	O
means	O
that	O
EFDT	O
can	O
deploy	O
useful	O
trees	O
much	O
sooner	O
than	O
VFDT	O
.	O
</s>
