Discrimination and visualization of different observed classes of edge-localized plasma instabilities (ELMs), using advanced data analysis techniques has been considered. An automated ELM type classifier which effectively incorporates measurement uncertainties is developed herein and applied to the discrimination of type I and type III ELMs in a set of carbon-wall JET plasmas. The approach involves constructing probability distribution functions (PDFs) for inter-ELM waiting times and global plasma parameters and then utilizing an effective similarity measure for comparing distributions: the Rao geodesic distance (GD). It is demonstrated that complete probability distributions of plasma parameters contain significantly more information than the measurement value alone, enabling effective discrimination of ELM types.