Data mining using example-based methods in oceanographic forecast models

The abstract presents a hybrid system that has proved capable of creating a prediction detailing the physical interactions occurring in a rapidly changing oceanographic environment. The aim of the system is to identify and forecast the thermal structure of the water ahead of an ongoing vessel. The work focuses on the development of a system for forecasting the behaviour of complex environments, in which the underling knowledge of the domain is not completely available, the rules governing the system are fuzzy and the sets of data samples are limited and incomplete. The paper presents a hybrid approach that combines the ability of a case based reasoning system (CBR) for Selecting previous similar situations and the generalising ability of artificial neural networks (ANN) to guide the adaptation stage of the case base reasoning system. The system has been successfully tested in the Atlantic Ocean in September 1997