Extraction de relations spatio-temporelles à partir de données environnementales et de la santé

TitreExtraction de relations spatio-temporelles à partir de données environnementales et de la santé
Publication TypeThesis
Year of Publication2013
AuthorsAlatrista-Salas, H
Academic DepartmentPPME - University of New Caledonia and LIRMM - University of Montpellier
DegreePhD
Number of Pages157
Date Published10/2013
Thesis TypeComputer Science
Mots-clésCorrelations research, Data exploration, Epidemiological detection systems., Geographic information, Spatiotemporal data mining
Abstract

Recently, thanks to the advanced technologies, (e.g. smartphones, sensors, etc.), large amounts of spatiotemporal data are now available. Commonly, a given spatiotemporal dataset contains a set of rows each of which presents spatial and temporal information of a happened event. The spatial information could be a city, a neighborhood, a river, a GPS location, etc. meanwhile temporal information is the date-time of the concerned event. Knowledge extraction from spatiotemporal data has been studied in many years for understanding the evolution or the spreading of phenomena in both temporal and spatio dimensions. However, there are still many challenging issues we need to deal with.
For instance, the dynamics of an infectious disease can be described as: (1) the interactions between humans; (2) the transmission vector as well as (3) some unrevealed spatiotemporal mechanisms involved in its spreading. In fact, the varying of one of these components can trigger changes the interaction scheme between the components and finally alter the behaviour of the whole system.
In my thesis, I will concern on proposing novel spatiotemporal data mining techniques to capture this phenomenon. More specifically, two generic methods of pattern mining are proposed: (1) the first one enables us to extract sequential patterns including spatial characteristics from the data; and (2) we propose a novel type of patterns called spatiosequential patterns which are used to express the evolution of a set of events in an area and its near environment. Our proposed approaches were tested on real datasets associated to two spatiotemporal phenomena: the spreading of pollution in rivers in France and the epidemiological monitoring of dengue in New Caledonia. In additional, two qualitative measures and a pattern visualization prototype are also supplied to assist the experts in the selection of meaningful patterns.