Percorrer por autor "Lage, Serge Gaspar Aguiar Fernandes"
A mostrar 1 - 1 de 1
Resultados por página
Opções de ordenação
- Learning portuguese fishing data patternsPublication . Lage, Serge Gaspar Aguiar Fernandes; Pinto, Iola Maria Silvério; Ferreira, João Carlos AmaroPortugal is a country historically linked to the sea with fishing being a very important activities for the Portuguese economy. On the other hand, tax fraud is present on fishery as well as in other economic activity and it is a harmful phenomenon for Portugal. For that reason there is a need to create ways to inspect this activity more efficiently. With the motivation to contribute to the resolution of this problem, the objectives of this dissertation are to analyze the data in order to derive patterns, which, when compared to real data can generate alerts for the existence of unusual activities. Concretely, the first objective seeks to infer when the vessels are fishing, and when they fish in an area other than the usual one, this using only velocity and location data. The second objective consists in classify the fishing license by taking into account the VMS data, more precisely velocity and position data. There are several studies developed in this area. What separates my work from the other studies found is the use of only data created by a device on board, in which there is no human interference. In the current developed solution the data is produced by the system MONICAP, a BLUEBOX system, mandatory for vessels over 12 meters in the European Union. This system records velocity, heading and location data. Concerning the first objective, a machine learning system, using velocity data, will be used to identify whether the vessel is fishing. The methodology used is based in clustering algorithms, to identify whether the fishing zone is usual or not. In addition the Hill Climbing algorithm and the Kernel density estimator are used to classify data as fishing or not. This system is designed so that it can be integrated into MONICAP itself. For the second objective, we will use data mining methods, as Random Forests, Neural Networks an others, to analyze possible associations between the data provided by MONICAP and the type of fishing license. The models were tested and evaluated using well-established data mining techniques following the procedures in Cross Industry Standard Process for Data Mining. The second solution allowed to show that it is possible to classify the fishing license by taking into account the VMS data, more precisely velocity and position data. The use of velocity and location turns out to be enough to create systems capable of satisfying the proposed objectives, so goals are all achieved.