PSI - Issue 64

6

Nikhil Holsamudrkar et al. / Procedia Structural Integrity 64 (2024) 580–587 Holsamudrkar Nikhil et al./ Structural Integrity Procedia 00 (2019) 000 – 000

585

Fig. 5. Training and validation accuracy evolution for each iteration

4. Results and Discussion The model shows a test accuracy of 87%. The confusion matrix and ROC curve, as shown in Fig. 6, are fundamental tools for judging its overall performance. As illustrated in Fig. 6 – (a), each matrix cell delineates the count of instances where the model's predictions correspond to the true labels, with diagonal elements denoting accurate classifications and off-diagonal elements indicating misclassifications. Analysis of the confusion matrix reveals strong performance in accurately identifying instances of steel yielding due to unique low amplitude signal characteristics. Other modes show about 84% true positives.

Fig. 6 (a) Confusion matrix expressed as a percentage, and (b) ROC curve.