Artificial Neural Networks Applied to the Classification of Emission Lines in Inductively Coupled Plasma–Atomic Emission Spectroscopy

Abstract Spectral-analytical lines used in inductively coupled plasma–atomic emission spectroscopy were classified in two different groups, identified as soft and hard lines, using artificial neural networks. The neural strategies of back-propagation of errors and self-organizing maps were applied. The spectral lines were characterized by means of their theoretical spectral data, namely wavelength, the excitation energies and the lower level of the electronic transition, statistical weights of the upper and lower level and the transition probability. The analytical sensitivity of the lines was also considered. The prediction results with the back-propagation model were obtained with an error lower than 0.1% in the test set. The self-organizing map allowed the localization of unknown lines in two different regions of a Kohonen map, which were identified as soft, and hard lines. The results were obtained with a similarity of 97% with respect to the back-propagation method.