PSI - Issue 41

8

America Califano et al. / Procedia Structural Integrity 41 (2022) 145–157 Author name / Structural Integrity Procedia 00 (2019) 000–000

152

Figure 5 - Separation Lines (SL) in plane z = 6 selected within the BV for each configuration. The green and red points follow the logic of the SED criterion previously explained.

Among the dozens of machine learning classification algorithms, an attempt has been done by using the eXtreme Gradient Boosting (XGBoost) machines (Chen and Guestrin (2016)). They have recently won several machine learning competitions, proving to be highly reliable, extremely scalable and adaptable to many kinds of dataset. They are based on enhanced stochastic gradient boosted trees (Friedman (1999)) and they work well if applied to non-linear problems, like the one in object. Moreover, due to their scalability, they generally show promising results even with low-dimensional datasets. For evaluating the performances of the XGBoost machines in classification tasks, many scores may be evaluated and adopted. In this case, the goodness of the classification has been evaluated by means of the Receiver Operating Characteristic (ROC) curve, together with its Area Under the Curve (AUC) (Fawcett (2005)), and the False Negative Rate (FNR).

Figure 6 - Slope of the Separation Lines (SLs) with respect to the � � � � parameter for: configurations at � � 1.� (a), configurations at � � 1.5 (b) and configurations at � � �.� (c).