84
M. C. Ben Nasr et al.
detected positives from the actual positive in other terms T P R measures how
sensitive your model is to the positive class.
T P R
i =
true positives
true positives + f alse negatives
,
(3)
where i corresponds to the class (activity) of the subject (i = 1..7). The terms of
the confusion matrix presented in Fig. 4 are defined as follows:
Conf tpr (i, j) =
M ij
C
j=1 M ij
,
(4)
where C is the number of classes, M ij is the number of predictions of class i that
actually belongs to class j it is usually measured by comparing the test results
to the ground truth.
Positive Predictive Value: The proportion of the predictions made that are
actually true and happened. P P V Highlights mostly how refined our model is
and how frequent we have false alerts.
P P V
i =
true positives
true positives + f alse positives
.
(5)
The terms of the confusion matrix presented in Fig. 5 are defined as follows:
Conf ppv (i, j) =
M ij
C
i=1 M ij
,
(6)
3 Experimental Results
3.1 Feature Illustration
This section illustrates the feature analysis and interpretation by providing the
means of the SF M and SC for each activity.
The BCG signal we are working with is the same displayed on Fig. 1. The
mean values are given in Table 1. We note that the normal breathing mean value
of the SF M is the lowest which confirms the periodicity hypothesis. The values
of SF M and SC taken during the coughing portion (C2) as well as the movement
portion (C6) are relatively high which further confirms the non-periodicity in the
corresponding portions.
For the post-cough breathing, we notice that, unlike the portion of normal
breathing, the values of the descriptors are high and close to those during the
movement and coughing activities which supports our choice to isolate these
portions.
M. C. Ben Nasr et al.
detected positives from the actual positive in other terms T P R measures how
sensitive your model is to the positive class.
T P R
i =
true positives
true positives + f alse negatives
,
(3)
where i corresponds to the class (activity) of the subject (i = 1..7). The terms of
the confusion matrix presented in Fig. 4 are defined as follows:
Conf tpr (i, j) =
M ij
C
j=1 M ij
,
(4)
where C is the number of classes, M ij is the number of predictions of class i that
actually belongs to class j it is usually measured by comparing the test results
to the ground truth.
Positive Predictive Value: The proportion of the predictions made that are
actually true and happened. P P V Highlights mostly how refined our model is
and how frequent we have false alerts.
P P V
i =
true positives
true positives + f alse positives
.
(5)
The terms of the confusion matrix presented in Fig. 5 are defined as follows:
Conf ppv (i, j) =
M ij
C
i=1 M ij
,
(6)
3 Experimental Results
3.1 Feature Illustration
This section illustrates the feature analysis and interpretation by providing the
means of the SF M and SC for each activity.
The BCG signal we are working with is the same displayed on Fig. 1. The
mean values are given in Table 1. We note that the normal breathing mean value
of the SF M is the lowest which confirms the periodicity hypothesis. The values
of SF M and SC taken during the coughing portion (C2) as well as the movement
portion (C6) are relatively high which further confirms the non-periodicity in the
corresponding portions.
For the post-cough breathing, we notice that, unlike the portion of normal
breathing, the values of the descriptors are high and close to those during the
movement and coughing activities which supports our choice to isolate these
portions.