Summary
Modern machine learning approaches to multivariate geochemical tectonic discrimination overcome limitations of classical graphical methods, and leverage large public geochemical databases to achieve high classification accuracy. Here we document practical approaches to developing classification models using free and open source tools, and investigate methods for visualising classification uncertainties and high-dimensional spatial relationships. In applying these classification methods into deep time, issues of biases and secular change become apparent. We discuss some opportunities to build upon these models and provide richer insight for future classification models.