<s>
In	O
data	B-General_Concept
analysis	I-General_Concept
,	O
anomaly	B-Algorithm
detection	I-Algorithm
(	O
also	O
referred	O
to	O
as	O
outlier	B-Algorithm
detection	I-Algorithm
and	O
sometimes	O
as	O
novelty	B-Algorithm
detection	I-Algorithm
)	O
is	O
generally	O
understood	O
to	O
be	O
the	O
identification	O
of	O
rare	O
items	O
,	O
events	O
or	O
observations	O
which	O
deviate	O
significantly	O
from	O
the	O
majority	O
of	O
the	O
data	O
and	O
do	O
not	O
conform	O
to	O
a	O
well	O
defined	O
notion	O
of	O
normal	O
behaviour	O
.	O
</s>
<s>
Anomaly	B-Algorithm
detection	I-Algorithm
finds	O
application	O
in	O
many	O
domains	O
including	O
cyber	O
security	O
,	O
medicine	O
,	O
machine	B-General_Concept
vision	I-General_Concept
,	O
statistics	O
,	O
neuroscience	O
,	O
law	O
enforcement	O
and	O
financial	O
fraud	O
to	O
name	O
only	O
a	O
few	O
.	O
</s>
<s>
Three	O
broad	O
categories	O
of	O
anomaly	B-Algorithm
detection	I-Algorithm
techniques	O
exist	O
.	O
</s>
<s>
Supervised	B-Algorithm
anomaly	I-Algorithm
detection	I-Algorithm
techniques	O
require	O
a	O
data	O
set	O
that	O
has	O
been	O
labeled	O
as	O
"	O
normal	O
"	O
and	O
"	O
abnormal	O
"	O
and	O
involves	O
training	O
a	O
classifier	O
.	O
</s>
<s>
However	O
,	O
this	O
approach	O
is	O
rarely	O
used	O
in	O
anomaly	B-Algorithm
detection	I-Algorithm
due	O
to	O
the	O
general	O
unavailability	O
of	O
labelled	O
data	O
and	O
the	O
inherent	O
unbalanced	O
nature	O
of	O
the	O
classes	O
.	O
</s>
<s>
Unsupervised	B-Algorithm
anomaly	I-Algorithm
detection	I-Algorithm
techniques	O
assume	O
the	O
data	O
is	O
unlabelled	O
and	O
are	O
by	O
far	O
the	O
most	O
commonly	O
used	O
due	O
to	O
their	O
wider	O
and	O
relevant	O
application	O
.	O
</s>
<s>
Anomaly	B-Algorithm
detection	I-Algorithm
is	O
applicable	O
in	O
a	O
very	O
large	O
number	O
and	O
variety	O
of	O
domains	O
,	O
and	O
is	O
an	O
important	O
subarea	O
of	O
unsupervised	O
machine	O
learning	O
.	O
</s>
<s>
As	O
such	O
it	O
has	O
applications	O
in	O
cyber-security	O
intrusion	O
detection	O
,	O
fraud	O
detection	O
,	O
fault	O
detection	O
,	O
system	O
health	O
monitoring	O
,	O
event	O
detection	O
in	O
sensor	O
networks	O
,	O
detecting	O
ecosystem	O
disturbances	O
,	O
defect	O
detection	O
in	O
images	O
using	O
machine	B-General_Concept
vision	I-General_Concept
,	O
medical	O
diagnosis	O
and	O
law	O
enforcement	O
.	O
</s>
<s>
Anomaly	B-Algorithm
detection	I-Algorithm
was	O
proposed	O
for	O
intrusion	O
detection	O
systems	O
(	O
IDS	O
)	O
by	O
Dorothy	O
Denning	O
in	O
1986	O
.	O
</s>
<s>
Anomaly	B-Algorithm
detection	I-Algorithm
for	O
IDS	O
is	O
normally	O
accomplished	O
with	O
thresholds	O
and	O
statistics	O
,	O
but	O
can	O
also	O
be	O
done	O
with	O
soft	B-General_Concept
computing	I-General_Concept
,	O
and	O
inductive	O
learning	O
.	O
</s>
<s>
The	O
counterpart	O
of	O
anomaly	B-Algorithm
detection	I-Algorithm
in	O
intrusion	O
detection	O
is	O
misuse	O
detection	O
.	O
</s>
<s>
It	O
is	O
often	O
used	O
in	O
preprocessing	B-General_Concept
to	O
remove	O
anomalous	O
data	O
from	O
the	O
dataset	O
.	O
</s>
<s>
In	O
supervised	B-General_Concept
learning	I-General_Concept
,	O
removing	O
the	O
anomalous	O
data	O
from	O
the	O
dataset	O
often	O
results	O
in	O
a	O
statistically	O
significant	O
increase	O
in	O
accuracy	O
.	O
</s>
<s>
Many	O
anomaly	B-Algorithm
detection	I-Algorithm
techniques	O
have	O
been	O
proposed	O
in	O
literature	O
.	O
</s>
<s>
ELKI	B-Language
is	O
an	O
open-source	O
Java	O
data	O
mining	O
toolkit	O
that	O
contains	O
several	O
anomaly	B-Algorithm
detection	I-Algorithm
algorithms	O
,	O
as	O
well	O
as	O
index	O
acceleration	O
for	O
them	O
.	O
</s>
<s>
PyOD	O
is	O
an	O
open-source	O
Python	O
library	O
developed	O
specifically	O
for	O
anomaly	B-Algorithm
detection	I-Algorithm
.	O
</s>
<s>
scikit-learn	B-Application
is	O
an	O
open-source	O
Python	O
library	O
that	O
has	O
built	O
functionality	O
to	O
provide	O
unsupervised	B-Algorithm
anomaly	I-Algorithm
detection	I-Algorithm
.	O
</s>
<s>
–	O
ODDS	O
:	O
A	O
large	O
collection	O
of	O
publicly	O
available	O
outlier	B-Algorithm
detection	I-Algorithm
datasets	O
with	O
ground	B-General_Concept
truth	I-General_Concept
in	O
different	O
domains	O
.	O
</s>
<s>
at	O
Harvard	O
Dataverse	O
:	O
Datasets	O
for	O
Unsupervised	B-Algorithm
Anomaly	I-Algorithm
Detection	I-Algorithm
with	O
ground	B-General_Concept
truth	I-General_Concept
.	O
</s>
<s>
at	O
Research	O
Data	O
Australia	O
having	O
more	O
than	O
12,000	O
anomaly	B-Algorithm
detection	I-Algorithm
datasets	O
with	O
ground	B-General_Concept
truth	I-General_Concept
.	O
</s>
