Performance metrics values
Performance metrics values 6L6n8
We can call the values in Figure 21-A the positive values, since the higher these values, the better the performance of the pre-trained network and its higher ability to classify correctly, and vice versa.
Whereas, the values in Figure 21-B can be called the negative values, where the more these values decrease and approach zero, the better the performance of the pre-trained network, and its classifier tends to be perfect, and therefore the wrong classification is very low or almost non-existent, and vice versa.
From Figure 21-A and Figure 21-B that show the values of the performance metrics for five the pre-trained networks(MobileNetV2, AlexNet, GoogleNet, ResNet 18, and ResNet 50) used in this study. All results for all pre-trained networks that used in the current study scored high results but we can see that the best values of the performance metrics were recorded by ResNet 50+ TL+KFCV, where average positive values which included in Figure 21-A (98. 59% Accuracy, 97. 9% Sensitivity, 98. 9% Specificity, 97. 9% Precision, 97. 9 % F1-Score, and 98. 9% Negative predicted value).
Thus, the ResNet50 recorded the highest positive values listed in Figure 21-A among all the pre-trained networks used in this research.
Also, we can see from average negative values which included in Figure 21-B( 1. 4%Error, 1. 05% False positive rate, 2. 1% False negative rate, 2. 1%False discovery rate).
Thus, the ResNet50 recorded the lowest negative values listed in Figure 21-B among all the pre-trained networks used in this research.
We can call the values in
Figure
21-A the
positive
values, since the higher these values, the better the
performance
of the pre-trained
network
and its higher ability to classify
correctly
, and vice versa.
Whereas, the values in
Figure
21-B can
be called
the
negative
values, where the more these values decrease and approach zero, the better the
performance
of the pre-trained
network
, and its classifier tends to be perfect, and
therefore
the
wrong
classification is
very
low or almost non-existent, and vice versa.
From
Figure
21-A and
Figure
21-B that
show
the values of the
performance
metrics for five the pre-trained networks(MobileNetV2,
AlexNet
,
GoogleNet
,
ResNet
18, and
ResNet
50)
used
in this study. All results for all pre-trained
networks
that
used
in the
current
study scored high results
but
we can
see
that the best values of the
performance
metrics
were recorded
by
ResNet
50+ TL+
KFCV
, where average
positive
values which included in
Figure
21-A (98. 59% Accuracy, 97. 9% Sensitivity, 98. 9% Specificity, 97.
9%
Precision, 97. 9 % F1-Score, and 98. 9%
Negative
predicted value).
Thus
, the ResNet50 recorded the highest
positive
values listed in
Figure
21-A among all the pre-trained
networks
used
in this research.
Also
, we can
see
from average
negative
values which included in
Figure
21-B
(
1. 4%Error, 1. 05% False
positive
rate, 2. 1% False
negative
rate, 2. 1%False discovery rate).
Thus
, the ResNet50 recorded the lowest
negative
values listed in
Figure
21-B among all the pre-trained
networks
used
in this research.