The use of an artificial neural network to improve precision in trace level, quantitative analysis of heavy metal pollutants

The author has used various neural networks to process the response obtained from an electroanalytical technique used for the analysis of trace metal pollutants in liquids. A previous paper by H.S. Manwaring (1994), compared the capabilities of the GRNN and MLP in this respect. It is shown that using the neural network to make predictions of unknown sample concentrations shows an improvement, by a factor of about two, on the mean absolute error and the prediction confidence when compared with a traditional, calibration curve technique. In addition, the neural network method is shown to produce reliable predictions even with instrumental responses that are completely unsuitable for traditional processing