Information Technology Reference
In-Depth Information
This measure gives an indication of how good the overall performance of a clas-
sifier is. Moreover, we can also use the
error
measure for expressing the opposite.
It denotes the deviation of the measurement from the truth and it can be obtained in
terms of the accuracy:
error
=
1
−
accuracy
(2.21)
Sensitivity
, also known as the
true positive rate
or
recall
, is a measure of how
good is the classifier to correctly predict actual positives samples. Its formulation is:
TP
sensitivity
=
(2.22)
TP
+
FN
In contrast, the
specificity
measure, also called the
true negative rate
,showsthe
ability to correctly predict actual negative samples. It is formulated as:
TN
TN
specificity
=
(2.23)
+
FP
Themeasures above are the oneswemostly use throughout thiswork. Nevertheless
there are also some statistical measures that are widely used such as
precision
,also
known as
positive predictive value
, which is the rate of TP with respect to all the
predicted positives, and
F
1
-score
(or
F-measure
) which is a general measure of
the classifier's accuracy that combines
precision
an
sensitivity
. These two can be
estimated in the following way:
TP
precision
=
(2.24)
TP
+
FP
precision
·
sensitivity
F
1
-
score
=
2
·
(2.25)
precision
+
sensitivity
2.5.4.3 Qualitative Criteria
In order to select the most appropriate classification system for a particular appli-
cation, it is also interesting to take into account a set of various qualitative criteria
in addition to the statistical measures. These aspects can help to make important
trade-off decisions during this selection process. Some of them are described as
follows:
•
Online capability
: tells whether the system is able to perform activity recognition
in real-time.
•
Recognition time
: is the time delay associated with the activity prediction process.
For instance, the length of the time window related to each prediction and its CPU
processing time.
Search WWH ::
Custom Search