Carmelo Cassisi

Enhancing density-based clustering: Parameter reduction and outlier detection

Clustering is a widely used unsupervised data mining technique. It allows to identify structures in collections of objects by grouping them into classes, named clusters, in such a way that similarity of objects within any cluster is maximized and similarity of objects belonging to different clusters is minimized. In density-based clustering, a cluster is defined as a connected dense component and grows in the direction driven by the density. The basic structure of density-based clustering presents some common drawbacks: (i) parameters have to be set; (ii) the behavior of the algorithm is sensitive to the density of the starting object; and (iii) adjacent clusters of different densities could not be properly identified. In this paper, we address all the above problems. Our method, based on the concept of space stratification, efficiently identifies the different densities in the dataset and, accordingly, ranks the objects of the original space. Next, it exploits such a knowledge by projecting the original data into a space with one more dimension. It performs a density based clustering taking into account the reverse-nearest-neighbor of the objects. Our method also reduces the number of input parameters by giving a guideline to set them in a suitable way. Experimental results indicate that our algorithm is able to deal with clusters of different densities and outperforms the most popular algorithms DBSCAN and OPTICS in all the standard benchmark datasets.

Monitoring Seismo-volcanic and Infrasonic Signals at Volcanoes: Mt. Etna Case Study

Volcanoes generate a broad range of seismo-volcanic and infrasonic signals, whose features and variations are often closely related to volcanic activity. The study of these signals is hence very useful in the monitoring and investigation of volcano dynamics. The analysis of seismo-volcanic and infrasonic signals requires specifically developed techniques due to their unique characteristics, which are generally quite distinct compared with tectonic and volcano-tectonic earthquakes. In this work, we describe analysis methods used to detect and locate seismo-volcanic and infrasonic signals at Mt. Etna. Volcanic tremor sources are located using a method based on spatial seismic amplitude distribution, assuming propagation in a homogeneous medium. The tremor source is found by calculating the goodness of the linear regression fit (R2) of the log-linearized equation of the seismic amplitude decay with distance. The location method for long-period events is based on the joint computation of semblance and R2 values, and the location method of very long-period events is based on the application of radial semblance. Infrasonic events and tremor are located by semblance–brightness- and semblance-based methods, respectively. The techniques described here can also be applied to other volcanoes and do not require particular network geometries (such as arrays) but rather simple sparse networks. Using the source locations of all the considered signals, we were able to reconstruct the shallow plumbing system (above sea level) during 2011.

Similarity Measures and Dimensionality Reduction Techniques for Time Series Data Mining

© 2012 Cassisi et al., licensee InTech. This is an open access chapter distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Similarity Measures and Dimensionality Reduction Techniques for Time Series Data Mining

Probabilistic Reasoning Over Seismic Time Series: Volcano Monitoring by Hidden Markov Models at Mt. Etna

From January 2011 to December 2015, Mt. Etna was mainly characterized by a cyclic eruptive behavior with more than 40 lava fountains from New South-East Crater. Using the RMS (Root Mean Square) of the seismic signal recorded by stations close to the summit area, an automatic recognition of the different states of volcanic activity (QUIET, PRE-FOUNTAIN, FOUNTAIN, POST-FOUNTAIN) has been applied for monitoring purposes. Since values of the RMS time series calculated on the seismic signal are generated from a stochastic process, we can try to model the system generating its sampled values, assumed to be a Markov process, using Hidden Markov Models (HMMs). HMMs analysis seeks to recover the sequence of hidden states from the observations. In our framework, observations are characters generated by the Symbolic Aggregate approXimation (SAX) technique, which maps RMS time series values with symbols of a pre-defined alphabet. The main advantages of the proposed framework, based on HMMs and SAX, with respect to other automatic systems applied on seismic signals at Mt. Etna, are the use of multiple stations and static thresholds to well characterize the volcano states. Its application on a wide seismic dataset of Etna volcano shows the possibility to guess the volcano states. The experimental results show that, in most of the cases, we detected lava fountains in advance.

Interplay between Tectonics and Mount Etna’s Volcanism: Insights into the Geometry of the Plumbing System

Volcanoes are geologic manifestations of highly dynamic and complexly coupled physical and chemical processes in the interior of the Earth. Most volcanism on Earth occurs at plate boundaries in places where tectonic plates move apart (e.g. Iceland) and in places where tectonic plates come together with one plate plunging (subducting) below the other into the mantle (e.g. Pacific ring of fire). Conversely, intraplate volcanism is a type of volcanism occurring far from plate boundaries and whose origins are rather controversial. To know the working mode of a volcano in a given region it is necessary to understand the interplay between tectonics, deformation processes and magma transport through the lithosphere (e.g. Vigneresse, 1999; Petford et al., 2000). Deformation-induced fault-fracture networks have been regarded as efficient pathways through which magma is transported, stored and eventually erupted at the Earth’s surface (e.g. Clemens and Mawer, 1992; Petford et al., 2000). At active volcanoes, magmas rise toward the surface and can stagnate at different levels in the lithosphere, giving rise to magma bodies of different shape and size (Marsh, 2000). Nearly all volcanic eruptions are supplied with magma through dykes and inclined sheets whose initiation and eventual propagation to the surface or, alternatively, arrest at some depth in the volcano, depend on the stress state in the volcano (Gudmundsson, 2006). At the surface of active volcanic edifices, the majority of eruptive fissures have a radial configuration and tangential or oblique fissures are rare. However, within many eroded volcanic edifices, dykes and dyke-fed eruptive fissures commonly have more complex patterns, resulting from regional stresses, magmatic reservoirs, anisotropies or variations in topography (Acocella et al., 2009). Geophysics can provide information on the geometry of plumbing system and magma chambers, as well as on the mechanisms of emplacement of dykes. Among the different branches of geophysics, seismology is the most powerful tool to obtain information about

Motif Discovery on Seismic Amplitude Time Series: The Case Study of Mt Etna 2011 Eruptive Activity

Algorithms searching for similar patterns are widely used in seismology both when the waveforms of the events of interest are known and when there is no a priori-knowledge. Such methods usually make use of the cross-correlation coefficient as a measure of similarity; if there is no a-priori knowledge, they behave as brute-force searching algorithms. The disadvantage of these methods, preventing or limiting their application to very large datasets, is computational complexity. The Mueen–Keogh (MK) algorithm overcomes this limitation by means of two optimization techniques—the early abandoning concept and space indexing. Here, we apply the MK algorithm to amplitude time series retrieved from seismic signals recorded during episodic eruptive activity of Mt Etna in 2011. By adequately tuning the input to the MK algorithm we found eight motif groups characterized by distinct seismic amplitude trends, each related to a different phenomenon. In particular, we observed that earthquakes are accompanied by sharp increases and decreases in seismic amplitude whereas lava fountains are accompanied by slower changes. These results demonstrate that the MK algorithm, because of its particular features, may have wide applicability in seismology.

DBStrata: a system for density-based clustering and outlier detection based on stratification

Clustering is a widely used unsupervised data mining technique. In density-based clustering, a cluster is defined as a connected dense component and grows in the direction set by the density. In this paper we present a software system called DBStrata that implements the density-based clustering architecture together with several extensions able to boost the clustering performances and to efficiently identify outliers.