Respiratory Activity Classification
83
Spectral Centroid (SC): The SC indicates where the center of mass of the
spectrum is located.
SC(m) =
N
k=1 (|X(m, k)| ∗ f (k))
N
k=1 |X(m, k)|
,
(2)
where f (k) is the frequency in Hz of the bin k.
BCG Feature Engineering: The SFM and SC measures have been evaluated
on each frame of the BCG signal. However, in order to avoid complexity that
comes with it (overlap noise propagation, presence of different labels in the same
frame, frames too small to be representative...), we propose to create a time-series
out of the SFM and SC values in each frame.
– Signal decomposition: the original BCG signal is decomposed into frames of
length 1024 with an overlap of 960 samples. Hence we used a windowing
function. In this work we used a Hamming window with an increment of 64
samples.
– The feature vector F (m) = [SF M (m), SC(m)]
T is extracted from each frame.
– Each feature of raw data SF M = [sf m(1), sfm(2)...sf m(L)]
T (L is the number of frames) is transformed into a time-series (equivalent to a signal) by
overlapping and adding the sf m of each frame. Note that the latter is a
constant vector, whose value is sf m and whose length is the frame size.
The sf m signal and sc signal are then used for the purpose of classification.
2.4 Activities Classification
Respiratory activities classification has been performed using a K-nearest neighbors classifier. It is a non-parametric classification method which classifies a
sample based on a plurality vote of its neighbors. The sample is assigned to the
class most common among its K nearest neighbors (K is a positive integer) in
terms of minimal distance. The algorithm adopted is Fine KN N which is the
finest variation of KN N since it labels the new input with the same label as only
one of its nearest neighbour K = 1. The evaluation of the algorithm as well as
the classification results were conducted using k-fold cross validation with k = 5.
2.5 Classification Evaluation
The classification performance is evaluated in terms of true positive rate and
positive predicted value.
True Positive Rate: The performance of our model will mainly be measured
using the confusion matrix [12]. Specifically T P R measures the proportion of
83
Spectral Centroid (SC): The SC indicates where the center of mass of the
spectrum is located.
SC(m) =
N
k=1 (|X(m, k)| ∗ f (k))
N
k=1 |X(m, k)|
,
(2)
where f (k) is the frequency in Hz of the bin k.
BCG Feature Engineering: The SFM and SC measures have been evaluated
on each frame of the BCG signal. However, in order to avoid complexity that
comes with it (overlap noise propagation, presence of different labels in the same
frame, frames too small to be representative...), we propose to create a time-series
out of the SFM and SC values in each frame.
– Signal decomposition: the original BCG signal is decomposed into frames of
length 1024 with an overlap of 960 samples. Hence we used a windowing
function. In this work we used a Hamming window with an increment of 64
samples.
– The feature vector F (m) = [SF M (m), SC(m)]
T is extracted from each frame.
– Each feature of raw data SF M = [sf m(1), sfm(2)...sf m(L)]
T (L is the number of frames) is transformed into a time-series (equivalent to a signal) by
overlapping and adding the sf m of each frame. Note that the latter is a
constant vector, whose value is sf m and whose length is the frame size.
The sf m signal and sc signal are then used for the purpose of classification.
2.4 Activities Classification
Respiratory activities classification has been performed using a K-nearest neighbors classifier. It is a non-parametric classification method which classifies a
sample based on a plurality vote of its neighbors. The sample is assigned to the
class most common among its K nearest neighbors (K is a positive integer) in
terms of minimal distance. The algorithm adopted is Fine KN N which is the
finest variation of KN N since it labels the new input with the same label as only
one of its nearest neighbour K = 1. The evaluation of the algorithm as well as
the classification results were conducted using k-fold cross validation with k = 5.
2.5 Classification Evaluation
The classification performance is evaluated in terms of true positive rate and
positive predicted value.
True Positive Rate: The performance of our model will mainly be measured
using the confusion matrix [12]. Specifically T P R measures the proportion of