PSI - Issue 38

A. Cugniere et al. / Procedia Structural Integrity 38 (2022) 168–181 A. Cugniere, O. Tusch and A. Mösenbacher / Structural Integrity Procedia 00 (2021) 000 – 000

174

7

Such configuration allows to compare the efficiency of global versus local anomaly detection algorithms. Using a dummy dataset similar to the one represented in Figure 1, the three semi-supervised approaches are compared in figure 7:

Fig. 7. Comparison between Anomaly Detection Algorithms for a dummy dataset

• The One-Class SVM algorithm considers the data as one big cluster and choose the data points that lie outside of this cluster as anomalies. Therefore, it seems to carry out a rather local anomaly detection. • The Isolation Forest algorithm detects several clusters and choose the datapoints that lie outside of these clusters as anomalies, which correspond to a local anomaly detection., • Finally, the Local Outlier Factor algorithm classifies the data using a different approach, since the smallest of the three recognizable clusters is here considered a global anomaly. Similarly, the unsupervised approach can be tested on the same dataset in figure 8:

Fig. 8. Clustering of a dummy dataset using DBSCAN