Gaussian mixture model for the unsupervised classification of AgCu nanoalloys based on the common neighbor analysis

Abstract

In this short communication we describe the results obtained from the application of the Gaussian mixture model, a popular unsupervised learning algorithm, to some modified data sets gained after the global optimizations of three different AgCu nanoalloys. In particular we highlight both positive and negative aspects of such an approach to this kind of data. We show indeed that thanks to the Common Neighbor Analysis we are still able to describe nanoalloys well enough to exploit a physically meaningful separation in different structural families, even with a very low-dimensional representation. On the other hand, we show that the imposition of an energy cutoff over the data set is a delicate matter since it forces us to find a tradeoff between having a large set of data and having clean data.