Prediction levels of heavy metals (Zn, Cu and Mn) in current Holocene deposits of the eastern part of the Mediterranean Moroccan margin (Alboran Sea)

The Alboran basin is part of the bético Rif chain and represents a point of exchange through the Strait of Gibraltar between the Atlantic Ocean to the west and the Algerian-Balearic Basin to the east.The purpose of this work is placed in the prediction of levels of heavy metals (Zn, Cu and Mn) in the Holocene to current deposits in the eastern part of the Moroccan Mediterranean margin of the Alboran sea, using the Artificial Neural Networks MLP-type RNA (Multi Layer Perceptron) non recurrent and supervised learning.Various tests of robustness as: Akaike Information Criteria, Root Mean Square Error, and Maximum Average Percentage Error, allow the choice of the architecture of the neural network.The backpropagation algorithm is used to determine the weights and biases of the neural network.Based learning and test consists of 50 samples (observations) of sediment analyzed at four sampling stations.The independent variables are sedimentological, mineralogical and geochemical parameters.These parameters are: bathymetry, depth levels in carrots,% of sand, the fine fraction <40 microns,% CaCO3,% illite, % Smectite and %Kaolinite + % chlorite.The dependent variables (to predict), which are three in number, are the heavy metal contents (Zn, Cu and Mn) of the sediment.A comparative study has been established between the neural prediction model MLP type and conventional statistical models that is multiple linear regression MLR.The performance of ANN-MLP model clearly shows themselves higher than those established by multiple linear regression MLR.