<s>
Isolation	B-General_Concept
Forest	I-General_Concept
is	O
an	O
algorithm	O
for	O
data	O
anomaly	B-Algorithm
detection	I-Algorithm
initially	O
developed	O
by	O
Fei	O
Tony	O
Liu	O
and	O
Zhi-Hua	O
Zhou	O
in	O
2008	O
.	O
</s>
<s>
Isolation	B-General_Concept
Forest	I-General_Concept
detects	O
anomalies	O
using	O
binary	O
trees	O
.	O
</s>
<s>
Isolation	B-General_Concept
Forest	I-General_Concept
splits	O
the	O
data	O
space	O
using	O
lines	O
that	O
are	O
orthogonal	B-Application
to	O
the	O
origin	O
and	O
assigns	O
higher	O
anomaly	O
scores	O
to	O
data	O
points	O
that	O
need	O
fewer	O
splits	O
to	O
be	O
isolated	O
.	O
</s>
<s>
The	O
figure	O
on	O
the	O
right	O
shows	O
an	O
application	O
of	O
the	O
Isolation	B-General_Concept
Forest	I-General_Concept
algorithm	O
to	O
the	O
waiting	O
time	O
between	O
eruptions	O
and	O
the	O
duration	O
of	O
the	O
eruption	O
of	O
the	O
Old	O
Faithful	O
geyser	O
in	O
Yellowstone	O
National	O
Park	O
.	O
</s>
<s>
The	O
Isolation	B-General_Concept
Forest	I-General_Concept
(	O
iForest	O
)	O
algorithm	O
was	O
initially	O
proposed	O
by	O
Fei	O
Tony	O
Liu	O
,	O
Kai	O
Ming	O
Ting	O
and	O
Zhi-Hua	O
Zhou	O
in	O
2008	O
.	O
</s>
<s>
The	O
premise	O
of	O
the	O
Isolation	B-General_Concept
Forest	I-General_Concept
algorithm	O
is	O
that	O
anomalous	O
data	O
points	O
are	O
easier	O
to	O
separate	O
from	O
the	O
rest	O
of	O
the	O
sample	O
.	O
</s>
<s>
One	O
of	O
the	O
main	O
reasons	O
for	O
swamping	O
is	O
the	O
presence	O
of	O
too	O
much	O
data	O
for	O
the	O
purpose	O
of	O
anomaly	B-Algorithm
detection	I-Algorithm
,	O
which	O
implies	O
one	O
possible	O
solution	O
to	O
the	O
problem	O
is	O
sub-sampling	O
.	O
</s>
<s>
Unfortunately	O
,	O
high-dimensional	O
data	O
also	O
affects	O
the	O
detection	O
performance	O
of	O
iForest	O
,	O
but	O
the	O
performance	O
can	O
be	O
vastly	O
improved	O
by	O
adding	O
a	O
features	O
selection	O
test	O
,	O
like	O
Kurtosis	B-Error_Name
,	O
to	O
reduce	O
the	O
dimensionality	O
of	O
the	O
sample	O
space	O
.	O
</s>
<s>
Anomaly	B-Algorithm
detection	I-Algorithm
with	O
Isolation	B-General_Concept
Forest	I-General_Concept
is	O
a	O
process	O
composed	O
of	O
two	O
main	O
stages	O
:	O
</s>
<s>
The	O
algorithm	O
for	O
computing	O
the	O
anomaly	O
score	O
of	O
a	O
data	O
point	O
is	O
based	O
on	O
the	O
observation	O
that	O
the	O
structure	O
of	O
iTrees	O
is	O
equivalent	O
to	O
that	O
of	O
Binary	B-Language
Search	I-Language
Trees	I-Language
(	O
BST	O
)	O
:	O
a	O
termination	O
to	O
an	O
external	O
node	O
of	O
the	O
iTree	O
corresponds	O
to	O
an	O
unsuccessful	O
search	O
in	O
the	O
BST	O
.	O
</s>
<s>
,	O
an	O
algorithm	O
that	O
detects	O
data-anomalies	O
using	O
binary	O
trees	O
written	O
in	O
R	B-Language
.	O
Released	O
by	O
the	O
paper	O
's	O
first	O
author	O
in	O
2009	O
.	O
</s>
<s>
-	O
A	O
Spark/Scala	O
implementation	O
,	O
created	O
by	O
from	O
the	O
LinkedIn	O
Anti-Abuse	O
AI	O
team	O
.	O
</s>
<s>
-	O
An	O
implementation	O
of	O
Isolation	B-General_Concept
Forest	I-General_Concept
for	O
Anomaly	B-Algorithm
Detection	I-Algorithm
by	O
H2O-3	O
.	O
</s>
<s>
implementation	O
in	O
R	B-Language
by	O
.	O
</s>
<s>
with	O
examples	O
in	O
scikit-learn	B-Application
.	O
</s>
<s>
-	O
A	O
distributed	O
implementation	O
in	O
Scala	O
and	O
Python	O
,	O
which	O
runs	O
on	O
Apache	B-Language
Spark	I-Language
.	O
</s>
<s>
Other	O
variations	O
of	O
Isolation	B-General_Concept
Forest	I-General_Concept
algorithm	O
implementations	O
:	O
</s>
<s>
–	O
An	O
implementation	O
of	O
Extended	O
Isolation	B-General_Concept
Forest	I-General_Concept
for	O
Anomaly	B-Algorithm
Detection	I-Algorithm
by	O
.	O
</s>
<s>
-	O
An	O
implementation	O
of	O
Extended	O
Isolation	B-General_Concept
Forest	I-General_Concept
for	O
Anomaly	B-Algorithm
Detection	I-Algorithm
by	O
H2O-3	O
.	O
</s>
<s>
-	O
An	O
implementation	O
of	O
Isolation	B-General_Concept
Forest	I-General_Concept
and	O
its	O
variations	O
by	O
David	O
Cortes	O
.	O
</s>
