Application of Self Organized Maps and Curvilinear Component Analysis to the Discrimination of the Vesuvius Seismic Signals

This paper reports on the unsupervised analysis of seismic signals recorded by four stations situated on the Vesuvius area in Naples, Italy. The dataset under examination is composed of earthquakes and false events like thunders, quarry blasts and man-made undersea explosions. The goal is to use these specific data for comparing the performance of three projection methods that are well known to be able to exploit structures and organizes data, providing a framework for understanding and interpreting the relationships between data items, and suggesting simple descriptions of these relationships. The three unsupervised techniques under examination are: Principal Component Analysis (PCA), which is linear, Self-Organizing Map (SOM) and Curvilinear Component Analysis (CCA), which are nonlinear. The results show that, among the above techniques, SOM can better visualize the complex set of high-dimensional data allowing to discover their intrinsic clusters structure and eventually discriminate the earthquakes from the false events either natural (thunder) or artificial (quarry blast and undersea explosions).