<s>
The	O
evaluation	B-General_Concept
of	I-General_Concept
binary	I-General_Concept
classifiers	I-General_Concept
compares	O
two	O
methods	O
of	O
assigning	O
a	O
binary	O
attribute	O
,	O
one	O
of	O
which	O
is	O
usually	O
a	O
standard	O
method	O
and	O
the	O
other	O
is	O
being	O
investigated	O
.	O
</s>
<s>
To	O
evaluate	O
a	O
classifier	O
,	O
one	O
compares	O
its	O
output	O
to	O
another	O
reference	O
classification	O
–	O
ideally	O
a	O
perfect	O
classification	O
,	O
but	O
in	O
practice	O
the	O
output	O
of	O
another	O
gold	O
standard	O
test	O
–	O
and	O
cross	B-Application
tabulates	I-Application
the	O
data	O
into	O
a	O
2×2	O
contingency	B-Application
table	I-Application
,	O
comparing	O
the	O
two	O
classifications	O
.	O
</s>
<s>
These	O
can	O
be	O
arranged	O
into	O
a	O
2×2	O
contingency	B-Application
table	I-Application
(	O
confusion	B-General_Concept
matrix	I-General_Concept
)	O
,	O
conventionally	O
with	O
the	O
test	O
result	O
on	O
the	O
vertical	O
axis	O
and	O
the	O
actual	O
condition	O
on	O
the	O
horizontal	O
axis	O
.	O
</s>
<s>
The	O
contingency	B-Application
table	I-Application
and	O
the	O
most	O
common	O
derived	O
ratios	O
are	O
summarized	O
below	O
;	O
see	O
sequel	O
for	O
details	O
.	O
</s>
<s>
The	O
relationship	O
between	O
sensitivity	O
and	O
specificity	O
,	O
as	O
well	O
as	O
the	O
performance	O
of	O
the	O
classifier	O
,	O
can	O
be	O
visualized	O
and	O
studied	O
using	O
the	O
Receiver	B-Algorithm
Operating	I-Algorithm
Characteristic	I-Algorithm
(	O
ROC	O
)	O
curve	O
.	O
</s>
<s>
Not	O
used	O
very	O
much	O
is	O
the	O
complementary	O
statistic	O
,	O
the	O
fraction	O
incorrect	O
(	O
FiC	O
)	O
:	O
FC	O
+	O
FiC	O
=	O
1	O
,	O
or	O
(	O
FP	O
+	O
FN	O
)	O
/	O
( TP	O
+	O
TN	O
+	O
FP	O
+	O
FN	O
)	O
–	O
this	O
is	O
the	O
sum	O
of	O
the	O
antidiagonal	B-Algorithm
,	O
divided	O
by	O
the	O
total	O
population	O
.	O
</s>
<s>
Another	O
useful	O
single	O
measure	O
is	O
"	O
area	O
under	O
the	O
ROC	B-Algorithm
curve	I-Algorithm
"	O
,	O
AUC	O
.	O
</s>
<s>
An	O
F-score	B-General_Concept
is	O
a	O
combination	O
of	O
the	O
precision	O
and	O
the	O
recall	O
,	O
providing	O
a	O
single	O
score	O
.	O
</s>
<s>
The	O
traditional	O
or	O
balanced	O
F-score	B-General_Concept
(	O
F1	B-General_Concept
score	I-General_Concept
)	O
is	O
the	O
harmonic	O
mean	O
of	O
precision	O
and	O
recall	O
:	O
</s>
<s>
F-scores	B-General_Concept
do	O
not	O
take	O
the	O
true	O
negative	O
rate	O
into	O
account	O
and	O
,	O
therefore	O
,	O
are	O
more	O
suited	O
to	O
information	B-Library
retrieval	I-Library
and	O
information	B-General_Concept
extraction	I-General_Concept
evaluation	O
where	O
the	O
true	O
negatives	O
are	O
innumerable	O
.	O
</s>
<s>
Instead	O
,	O
measures	O
such	O
as	O
the	O
phi	B-General_Concept
coefficient	I-General_Concept
,	O
Matthews	B-General_Concept
correlation	I-General_Concept
coefficient	I-General_Concept
,	O
informedness	B-General_Concept
or	O
Cohen	B-General_Concept
's	I-General_Concept
kappa	I-General_Concept
may	O
be	O
preferable	O
to	O
assess	O
the	O
performance	O
of	O
a	O
binary	O
classifier	O
.	O
</s>
<s>
As	O
a	O
correlation	O
coefficient	O
,	O
the	O
Matthews	B-General_Concept
correlation	I-General_Concept
coefficient	I-General_Concept
is	O
the	O
geometric	O
mean	O
of	O
the	O
regression	B-General_Concept
coefficients	I-General_Concept
of	O
the	O
problem	O
and	O
its	O
dual	O
.	O
</s>
<s>
The	O
component	O
regression	B-General_Concept
coefficients	I-General_Concept
of	O
the	O
Matthews	B-General_Concept
correlation	I-General_Concept
coefficient	I-General_Concept
are	O
markedness	O
(	O
deltap	O
)	O
and	O
informedness	B-General_Concept
(	O
Youden	B-General_Concept
's	I-General_Concept
J	I-General_Concept
statistic	I-General_Concept
or	O
deltap	O
 '	O
)	O
.	O
</s>
