accuracy results will be discussed in the next section. Then, more the number of
iterations is increasing; more classification results are going higher. Table 2 depicts all
used parameters for the ECG classification process.
3.2 ECG Classification Results
The propounded artificial neural network is presented with the confusion matrix, the
histogram error, and the curve ROC and training performance results.
3.2.1 Confusion Matrix
A confusion matrix is a crucial method to determine the performance classification of a
system since it divides the results into four classes such as True Positive (TP), True
Negative (TN), False Positive (FP) and False Negative (FN). Accordingly, in Fig. 8, FP
results are shown in the confusion matrix in the last row and the FN are depicted in the
last column. It summarizes the prediction results on a classification issue. Therefore, it
answers the problem of determination of the class of each signal. Indeed, it is depicted
by a size of n  n associated with a classifier showing the predicted and actual classification, where n is the number of different classes. Table 1 shows a confusion matrix
for n = 3. It gives us a sight not only into the errors being made by a classifier but more
importantly the types of errors that are being made. The classification accuracy alone
can be misleading where we have an unequal number of observations in each class or in
case of having more than two classes in the dataset. Calculating a confusion matrix
gives a better idea of what a classification model is getting right and what types of
errors it is making. Indeed, performance results are calculated by the following
equations presenting the sensitivity, the specificity and the accuracy.
Sensitivity True positive rate
ð
Þ¼
TP
TP þ FN
ð3Þ
Specificity False positive rate
ð
Þ¼
TN
TN þ FP
ð4Þ
Accuracy percent of all samples correctly classified
ð
Þ ¼
TN þ TP
TN þ TP þ FN þ FP
ð5Þ
The prediction results (Fig. 8) show an accuracy of 100% for the training process,
80% for the validation process and 73.3% for the testing process. These results presented a high precision of classification, which proves the robustness of the used
artificial neural network architecture. Indeed, there are no misclassified signals in the
training confusion matrix. However, in the validation confusion matrix in all confusion
matrix a value of 0% for misclassified signal is achieved. The green colour in the
confusion matrix represents the true positive classified ECG signals; however, the red
colour depicts the misclassified signals. The first class presents the atrial fibrillation
Deep Learning-Based Approach for Atrial Fibrillation Detection
107
iterations is increasing; more classification results are going higher. Table 2 depicts all
used parameters for the ECG classification process.
3.2 ECG Classification Results
The propounded artificial neural network is presented with the confusion matrix, the
histogram error, and the curve ROC and training performance results.
3.2.1 Confusion Matrix
A confusion matrix is a crucial method to determine the performance classification of a
system since it divides the results into four classes such as True Positive (TP), True
Negative (TN), False Positive (FP) and False Negative (FN). Accordingly, in Fig. 8, FP
results are shown in the confusion matrix in the last row and the FN are depicted in the
last column. It summarizes the prediction results on a classification issue. Therefore, it
answers the problem of determination of the class of each signal. Indeed, it is depicted
by a size of n  n associated with a classifier showing the predicted and actual classification, where n is the number of different classes. Table 1 shows a confusion matrix
for n = 3. It gives us a sight not only into the errors being made by a classifier but more
importantly the types of errors that are being made. The classification accuracy alone
can be misleading where we have an unequal number of observations in each class or in
case of having more than two classes in the dataset. Calculating a confusion matrix
gives a better idea of what a classification model is getting right and what types of
errors it is making. Indeed, performance results are calculated by the following
equations presenting the sensitivity, the specificity and the accuracy.
Sensitivity True positive rate
ð
Þ¼
TP
TP þ FN
ð3Þ
Specificity False positive rate
ð
Þ¼
TN
TN þ FP
ð4Þ
Accuracy percent of all samples correctly classified
ð
Þ ¼
TN þ TP
TN þ TP þ FN þ FP
ð5Þ
The prediction results (Fig. 8) show an accuracy of 100% for the training process,
80% for the validation process and 73.3% for the testing process. These results presented a high precision of classification, which proves the robustness of the used
artificial neural network architecture. Indeed, there are no misclassified signals in the
training confusion matrix. However, in the validation confusion matrix in all confusion
matrix a value of 0% for misclassified signal is achieved. The green colour in the
confusion matrix represents the true positive classified ECG signals; however, the red
colour depicts the misclassified signals. The first class presents the atrial fibrillation
Deep Learning-Based Approach for Atrial Fibrillation Detection
107