Training Data Parameters
Training parameters have a crucial role to obtain excellent accuracy. For this fact, the
number of parameters should be done with the exact precision to get high accuracy,
sensitivity and specificity results and to yield the best values. In fact, For ANN1, we
use 10 hidden layer with 10 neurons for each layer. Thus, we obtain a number of 100
parameters, presenting the feature maps extracted from ECG signals. However, for
ANN2 and ANN3, the number of hidden layers is 2 and 20 respectively. Thus,
Fig. 7. ANN trained architecture: (a) ANN with HL = 10, (b) ANN with HL = 2, (c) ANN with
HL = 20.
Table 2. Training data parameters
Training parameters ANN1
ANN2
ANN3
Iterations
1000
1000
1000
Activation function Sigmoid
Sigmoid
Sigmoid
Classifier function Soft-max
Soft-max
Soft-max
Hidden layers
10
2
20
Parameters numbers 100
20
200
Train-samples
142
142
142
Validation-samples 30
30
30
Test-images
30
30
30
Error-rate
0.001
0.001
0.001
Batch
3600 samples 3600 samples 3600 samples
106
L. Khriji et al.