“…In this approach, any pattern that stands out with respect to other data-points is considered as anomaly, paying however the necessary attention in discriminating anomalies from novelties in the observed behaviours (referred as problem of 'novelty detection', [12], [13], [14]). Various approaches involving different techniques, usually based on learning processes, have been used for discriminating anomalies from regular data points: classification of patterns by means of neural networks ( [15], [16]), Bayesian networks ( [17], [18]), SVM ( [19]) or rule-based systems ( [20]), clustering of data for outliers identification ( [21], [22]), distance or density analysis respect to nearest neighbour ( [23]), statistical approaches leveraging parametric models (Gaussian regression models, [24], [25]) or Kernel Functions ( [26]), information-theoretic techniques based on entropy ( [27]) or Kologomorov complexity ( [28]), spectral analysis performed, e.g., by means of Principal Components Analysis (PCA, [29] ) or wavelet transform ( [15]). There have been also interesting studies exploiting the recent developments in Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) for the anomaly detection, directly based on the images [30], [31], [32], [33].…”