A neural network approach to crystal structure classification

摘要

This paper focuses classification of crystal classes in a periodic table using the known neural network (NN) learning algorithm, viz. generalized delta rule (GDR) by fddeing the set of input features in max-min-max sub arrays. We have taken eighteen independent physical parameters for each element, trained the network from atomic number (AN) 1 to 84 and we validated the crystal class from AN 86 to 95 from the trained network and achieved 100 per cent accuracy, which was later expended from AN 96 to 120. Further, we have also evaluated the dependencies of the neural network in different confidence intervals and hidden layers. We would like to call this learning algorithm as max-min-max GDR.