<s>
An	O
alternating	B-Algorithm
decision	I-Algorithm
tree	I-Algorithm
(	O
ADTree	B-Algorithm
)	O
is	O
a	O
machine	O
learning	O
method	O
for	O
classification	O
.	O
</s>
<s>
It	O
generalizes	O
decision	B-Algorithm
trees	I-Algorithm
and	O
has	O
connections	O
to	O
boosting	B-Algorithm
.	O
</s>
<s>
An	O
ADTree	B-Algorithm
consists	O
of	O
an	O
alternation	O
of	O
decision	O
nodes	O
,	O
which	O
specify	O
a	O
predicate	O
condition	O
,	O
and	O
prediction	O
nodes	O
,	O
which	O
contain	O
a	O
single	O
number	O
.	O
</s>
<s>
An	O
instance	O
is	O
classified	O
by	O
an	O
ADTree	B-Algorithm
by	O
following	O
all	O
paths	O
for	O
which	O
all	O
decision	O
nodes	O
are	O
true	O
,	O
and	O
summing	O
any	O
prediction	O
nodes	O
that	O
are	O
traversed	O
.	O
</s>
<s>
ADTrees	B-Algorithm
were	O
introduced	O
by	O
Yoav	O
Freund	O
and	O
Llew	O
Mason	O
.	O
</s>
<s>
Implementations	O
are	O
available	O
in	O
Weka	B-Language
and	O
JBoost	O
.	O
</s>
<s>
or	O
decision	B-Algorithm
trees	I-Algorithm
as	O
weak	O
hypotheses	O
.	O
</s>
<s>
is	O
the	O
number	O
of	O
boosting	B-Algorithm
iterations	O
)	O
,	O
which	O
then	O
vote	O
on	O
the	O
final	O
classification	O
according	O
to	O
their	O
weights	O
.	O
</s>
<s>
Individual	O
decision	B-Algorithm
stumps	I-Algorithm
are	O
weighted	O
according	O
to	O
their	O
ability	O
to	O
classify	O
the	O
data	O
.	O
</s>
<s>
Boosting	B-Algorithm
a	O
simple	O
learner	O
results	O
in	O
an	O
unstructured	O
set	O
of	O
hypotheses	O
,	O
making	O
it	O
difficult	O
to	O
infer	O
correlations	O
between	O
attributes	O
.	O
</s>
<s>
Alternating	B-Algorithm
decision	I-Algorithm
trees	I-Algorithm
introduce	O
structure	O
to	O
the	O
set	O
of	O
hypotheses	O
by	O
requiring	O
that	O
they	O
build	O
off	O
a	O
hypothesis	O
that	O
was	O
produced	O
in	O
an	O
earlier	O
iteration	O
.	O
</s>
<s>
An	O
alternating	B-Algorithm
decision	I-Algorithm
tree	I-Algorithm
consists	O
of	O
decision	O
nodes	O
and	O
prediction	O
nodes	O
.	O
</s>
<s>
ADTrees	B-Algorithm
always	O
have	O
prediction	O
nodes	O
as	O
both	O
root	O
and	O
leaves	O
.	O
</s>
<s>
An	O
instance	O
is	O
classified	O
by	O
an	O
ADTree	B-Algorithm
by	O
following	O
all	O
paths	O
for	O
which	O
all	O
decision	O
nodes	O
are	O
true	O
and	O
summing	O
any	O
prediction	O
nodes	O
that	O
are	O
traversed	O
.	O
</s>
<s>
This	O
is	O
different	O
from	O
binary	O
classification	O
trees	O
such	O
as	O
CART	O
(	O
Classification	B-Algorithm
and	I-Algorithm
regression	I-Algorithm
tree	I-Algorithm
)	O
or	O
C4.5	B-Algorithm
in	O
which	O
an	O
instance	O
follows	O
only	O
one	O
path	O
through	O
the	O
tree	O
.	O
</s>
<s>
The	O
original	O
authors	O
list	O
three	O
potential	O
levels	O
of	O
interpretation	O
for	O
the	O
set	O
of	O
attributes	O
identified	O
by	O
an	O
ADTree	B-Algorithm
:	O
</s>
<s>
The	O
inputs	O
to	O
the	O
alternating	B-Algorithm
decision	I-Algorithm
tree	I-Algorithm
algorithm	O
are	O
:	O
</s>
<s>
The	O
fundamental	O
element	O
of	O
the	O
ADTree	B-Algorithm
algorithm	O
is	O
the	O
rule	O
.	O
</s>
<s>
Figure	O
6	O
in	O
the	O
original	O
paper	O
demonstrates	O
that	O
ADTrees	B-Algorithm
are	O
typically	O
as	O
robust	O
as	O
boosted	O
decision	B-Algorithm
trees	I-Algorithm
and	O
boosted	O
decision	B-Algorithm
stumps	I-Algorithm
.	O
</s>
