Uso de redes neurais para estimar o conteúdo de chumbo em diferentes amostras de chocolates

Abstract

The world production of cocoa is in the order of a few million tons per year and is restricted to tropical areas such as Brazil, Indonesia and the West Africa, with emphasis on the Ivory Coast and Ghana. The intensive use of pesticides and other agricultural pesticides, contaminated water, mining, air pollutants and soil composition are factors that can potentially incorporate potentially harmful concentrations of metals such as lead, cadmium and nickel into cocoa trees and their products. The presence of contaminants in chocolate mainly affects children, who are major consumers of the product, as they absorb these metals more easily and are also more vulnerable to contamination. From the databases available in the literature, a neural network capable of estimating the lead concentration in chocolate bars was developed with aluminum and copper concentrations as inputs. The network was designed using the MatLab R2015a Neural Network toolbox, and the results obtained had a coefficient of determination of 0.9760 for the 61 points used in the training and 0.9846 for the 6 verification points, a satisfactory performance both in adjustment and generalization. Through the calculated weights and biases, it is possible with few calculations to estimate lead concentrations for chocolate samples with known aluminum and copper concentrations, an application with great potential for financial and operational savings for the cocoa derivatives industry.