Performance of the DS model during training. The top figure shows the evolution of the loss function (Eq. (19)) for the training and validation set. The bottom figure shows the evolution of a number of classification metrics based on the number of true positives TP, true negatives TN, false positives FP, and false negatives FN, including: precision , recall , selectivity , accuracy , balanced accuracy (average of recall and selectivity), and F1-score (harmonic mean of recall and precision). After 165 epochs, the model has reached its maximum validation F1-score.