Application of the Artificial Neural Networks of MLP Type for the Prediction of the Levels of Heavy Metals in Moroccan Aquatic Sediments

The present work describes a new approach to the prediction of the concentrations of heavy metals in Moroccan river sediments relying on a number of physico-chemical parameters. The originality of this work lies in the application of neural networks the application of neural networks MLP type (Multilayer Perceptron). During the first step, the parameters of the neurons are determined by the method supervised. In a second step, the weights of output connections are determined using the algorithm of gradient back propagation. The data used as the basis for learning of the neuronal model are those related to the analysis of the sediment samples collected at the level of several stations, distributed in space and time, of the watershed of the river Beht of the region Khemisset in Morocco. The dependent variables (to explain or predict), which are three, are containing heavy metal (Cu, Pb and Cr) of sediments. A comparative study was established between the neuronal model for prediction of MLP type and conventional statistical models namely the MLR (Multiple linear regression).The performance of the predictive model established by the type MLP neural networks are significantly higher than those established by the multiple linear regression.