Applying neural networks with active neurons to sea-water quality measurements

E. Hatzikos, L. Anastasakis, N. Bassiliades, I. Vlahavas, “Applying neural networks with active neurons to sea-water quality measurements”, 2nd Int. Scientific Conf. on Computer Science, Plamenka Borovska, Sofoklis Christofordis (Ed.), IEEE Computer Society, Bulgarian Section, 30th Sep-2nd Oct 2005, Halkidiki, Greece, 2005.

2nd Int. Scientific Conf. on Computer Science, Plamenka Borovska, Sofoklis Christofordis (Ed.), IEEE Computer Society, Bulgarian Section, 30th Sep-2nd Oct 2005, Halkidiki, Greece, 2005.

This study examines the presence of either linear or nonlinear relationships between a number of popular sea-water quality indicators such as water temperature, pH, amount of dissolved oxygen and turbidity. The data are obtained from a set of sensors in an underwater measurement station. The neural networks with active neurons are applied to the prediction of each one of the above four indicators and their performance is compared against a benchmark prediction method known as the random walk model. The random walk model is the simpler prediction method, which accepts as the best prediction for a variable its current value. The neural network with active neurons is a black box method, which contrary to neural networks with passive neurons does not require a long set of training data. The results show that for daily predictions the neural network with active neurons is able to beat the random walk model with regard to directional accuracy, namely the direction (upward or downwards) of the modelling object in the next day.