<s>
Choosing	O
informative	O
,	O
discriminating	O
and	O
independent	O
features	B-Algorithm
is	O
a	O
crucial	O
element	O
of	O
effective	O
algorithms	O
in	O
pattern	O
recognition	O
,	O
classification	B-General_Concept
and	O
regression	O
.	O
</s>
<s>
Features	B-Algorithm
are	O
usually	O
numeric	O
,	O
but	O
structural	O
features	B-Algorithm
such	O
as	O
strings	O
and	O
graphs	O
are	O
used	O
in	O
syntactic	B-Algorithm
pattern	I-Algorithm
recognition	I-Algorithm
.	O
</s>
<s>
The	O
concept	O
of	O
"	O
feature	O
"	O
is	O
related	O
to	O
that	O
of	O
explanatory	O
variable	O
used	O
in	O
statistical	B-General_Concept
techniques	I-General_Concept
such	O
as	O
linear	B-General_Concept
regression	I-General_Concept
.	O
</s>
<s>
A	O
numeric	O
feature	O
can	O
be	O
conveniently	O
described	O
by	O
a	O
feature	B-Algorithm
vector	I-Algorithm
.	O
</s>
<s>
One	O
way	O
to	O
achieve	O
binary	B-General_Concept
classification	I-General_Concept
is	O
using	O
a	O
linear	B-General_Concept
predictor	I-General_Concept
function	I-General_Concept
(	O
related	O
to	O
the	O
perceptron	B-Algorithm
)	O
with	O
a	O
feature	B-Algorithm
vector	I-Algorithm
as	O
input	O
.	O
</s>
<s>
The	O
method	O
consists	O
of	O
calculating	O
the	O
scalar	O
product	O
between	O
the	O
feature	B-Algorithm
vector	I-Algorithm
and	O
a	O
vector	O
of	O
weights	O
,	O
qualifying	O
those	O
observations	O
whose	O
result	O
exceeds	O
a	O
threshold	O
.	O
</s>
<s>
Algorithms	O
for	O
classification	B-General_Concept
from	O
a	O
feature	B-Algorithm
vector	I-Algorithm
include	O
nearest	B-General_Concept
neighbor	I-General_Concept
classification	I-General_Concept
,	O
neural	B-Architecture
networks	I-Architecture
,	O
and	O
statistical	B-General_Concept
techniques	I-General_Concept
such	O
as	O
Bayesian	O
approaches	O
.	O
</s>
<s>
In	O
character	B-Application
recognition	I-Application
,	O
features	B-Algorithm
may	O
include	O
histograms	B-Algorithm
counting	O
the	O
number	O
of	O
black	O
pixels	O
along	O
horizontal	O
and	O
vertical	O
directions	O
,	O
number	O
of	O
internal	O
holes	O
,	O
stroke	O
detection	O
and	O
many	O
others	O
.	O
</s>
<s>
In	O
speech	B-Application
recognition	I-Application
,	O
features	B-Algorithm
for	O
recognizing	O
phonemes	B-Language
can	O
include	O
noise	O
ratios	O
,	O
length	O
of	O
sounds	O
,	O
relative	O
power	O
,	O
filter	O
matches	O
and	O
many	O
others	O
.	O
</s>
<s>
In	O
spam	O
detection	O
algorithms	O
,	O
features	B-Algorithm
may	O
include	O
the	O
presence	O
or	O
absence	O
of	O
certain	O
email	O
headers	O
,	O
</s>
<s>
In	O
computer	B-Application
vision	I-Application
,	O
there	O
are	O
a	O
large	O
number	O
of	O
possible	O
features	B-Algorithm
,	O
such	O
as	O
edges	O
and	O
objects	O
.	O
</s>
<s>
In	O
pattern	O
recognition	O
and	O
machine	O
learning	O
,	O
a	O
feature	B-Algorithm
vector	I-Algorithm
is	O
an	O
n-dimensional	O
vector	O
of	O
numerical	O
features	B-Algorithm
that	O
represent	O
some	O
object	O
.	O
</s>
<s>
When	O
representing	O
images	O
,	O
the	O
feature	O
values	O
might	O
correspond	O
to	O
the	O
pixels	O
of	O
an	O
image	O
,	O
while	O
when	O
representing	O
texts	O
the	O
features	B-Algorithm
might	O
be	O
the	O
frequencies	O
of	O
occurrence	O
of	O
textual	O
terms	O
.	O
</s>
<s>
Feature	B-Algorithm
vectors	I-Algorithm
are	O
equivalent	O
to	O
the	O
vectors	O
of	O
explanatory	O
variables	O
used	O
in	O
statistical	O
procedures	O
such	O
as	O
linear	B-General_Concept
regression	I-General_Concept
.	O
</s>
<s>
Feature	B-Algorithm
vectors	I-Algorithm
are	O
often	O
combined	O
with	O
weights	O
using	O
a	O
dot	O
product	O
in	O
order	O
to	O
construct	O
a	O
linear	B-General_Concept
predictor	I-General_Concept
function	I-General_Concept
that	O
is	O
used	O
to	O
determine	O
a	O
score	O
for	O
making	O
a	O
prediction	O
.	O
</s>
<s>
In	O
order	O
to	O
reduce	O
the	O
dimensionality	O
of	O
the	O
feature	O
space	O
,	O
a	O
number	O
of	O
dimensionality	B-Algorithm
reduction	I-Algorithm
techniques	O
can	O
be	O
employed	O
.	O
</s>
<s>
Higher-level	O
features	B-Algorithm
can	O
be	O
obtained	O
from	O
already	O
available	O
features	B-Algorithm
and	O
added	O
to	O
the	O
feature	B-Algorithm
vector	I-Algorithm
;	O
for	O
example	O
,	O
for	O
the	O
study	O
of	O
diseases	O
the	O
feature	O
'	O
Age	O
 '	O
is	O
useful	O
and	O
is	O
defined	O
as	O
Age	O
=	O
'	O
Year	O
of	O
death	O
 '	O
minus	O
'	O
Year	O
of	O
birth	O
 '	O
.	O
</s>
<s>
This	O
process	O
is	O
referred	O
to	O
as	O
feature	B-Algorithm
construction	I-Algorithm
.	O
</s>
<s>
Feature	B-Algorithm
construction	I-Algorithm
is	O
the	O
application	O
of	O
a	O
set	O
of	O
constructive	O
operators	O
to	O
a	O
set	O
of	O
existing	O
features	B-Algorithm
resulting	O
in	O
construction	O
of	O
new	O
features	B-Algorithm
.	O
</s>
<s>
Examples	O
of	O
such	O
constructive	O
operators	O
include	O
checking	O
for	O
the	O
equality	O
conditions	O
{=	O
,	O
≠}	O
,	O
the	O
arithmetic	O
operators	O
{+	O
,	O
−	O
,	O
×	O
,	O
/	O
}	O
,	O
the	O
array	O
operators	O
{	O
max(S )	O
,	O
min(S )	O
,	O
average(S )	O
}	O
as	O
well	O
as	O
other	O
more	O
sophisticated	O
operators	O
,	O
for	O
example	O
count(S,C )	O
that	O
counts	O
the	O
number	O
of	O
features	B-Algorithm
in	O
the	O
feature	B-Algorithm
vector	I-Algorithm
S	O
satisfying	O
some	O
condition	O
C	O
or	O
,	O
for	O
example	O
,	O
distances	O
to	O
other	O
recognition	O
classes	O
generalized	O
by	O
some	O
accepting	O
device	O
.	O
</s>
<s>
Feature	B-Algorithm
construction	I-Algorithm
has	O
long	O
been	O
considered	O
a	O
powerful	O
tool	O
for	O
increasing	O
both	O
accuracy	O
and	O
understanding	O
of	O
structure	O
,	O
particularly	O
in	O
high-dimensional	O
problems	O
.	O
</s>
<s>
The	O
initial	O
set	O
of	O
raw	O
features	B-Algorithm
can	O
be	O
redundant	O
and	O
too	O
large	O
to	O
be	O
managed	O
.	O
</s>
<s>
Therefore	O
,	O
a	O
preliminary	O
step	O
in	O
many	O
applications	O
of	O
machine	O
learning	O
and	O
pattern	O
recognition	O
consists	O
of	O
selecting	B-General_Concept
a	O
subset	O
of	O
features	B-Algorithm
,	O
or	O
constructing	B-Algorithm
a	O
new	O
and	O
reduced	O
set	O
of	O
features	B-Algorithm
to	O
facilitate	O
learning	O
,	O
and	O
to	O
improve	O
generalization	O
and	O
interpretability	O
.	O
</s>
<s>
Extracting	O
or	O
selecting	B-General_Concept
features	B-Algorithm
is	O
a	O
combination	O
of	O
art	O
and	O
science	O
;	O
developing	O
systems	O
to	O
do	O
so	O
is	O
known	O
as	O
feature	B-General_Concept
engineering	I-General_Concept
.	O
</s>
<s>
Automating	O
this	O
process	O
is	O
feature	B-General_Concept
learning	I-General_Concept
,	O
where	O
a	O
machine	O
not	O
only	O
uses	O
features	B-Algorithm
for	O
learning	O
,	O
but	O
learns	O
the	O
features	B-Algorithm
itself	O
.	O
</s>
