Training an artificial neural network for recognizing electron collision patterns (Q822558)

The aim here is to develop a classification model based on artificial neural networks to distinguish electron scattering cross section (ESCS) data into the collition types of: attachment, elastic, excitation and ionization. The ESCS are used in the domain of plasma modeling. The proposed artificial neural network (ANN) model is a classical multilayer perceptron where the weights and the bias are to be trained by a gradient descent algorithm, the so-called Back propagation algorithm. The non-trainable parameters as for instance the number of nodes in each layer, are optimized by a Bayesian optimization approach. The performance of the model has been checked by means of a Monte Carlo cross validation procedure. Regarding data processing, one reports that all the scattering cross section data for collisions between electron and molecules have been downloaded from the Plasma Data Exchange Project, LXCat, data that are available in a Supplementary Material. The third section of the article is devoted to the description of the Experiments, Statistical performance analysis and Feature investigation. A short discussion can be found in the fourth section. Supplementary Material related to this article is contained in: \url{https://doi.orf/10.5281/zenodo.4015194} and \url{https://doi.org/10.1016/j.physleta.2020.127005}.