A metrological framework for uncertainty evaluation in machine learning classification models

The introduction of a metrological framework for uncertainty evaluation in machine learning classification models marks a significant advancement in the field. As machine learning becomes increasingly integral to applications such as climate and earth observation, medical diagnosis, and bioaerosol monitoring, the need for reliable predictions accompanied by uncertainty assessments has never been more critical. Current standards, particularly the International Vocabulary of Metrology (VIM) and the Guide to the Expression of Uncertainty in Measurement (GUM), do not adequately address uncertainty evaluation for nominal properties, which are essential in these applications. The proposed framework, based on probability mass functions and summary statistics, aims to fill this gap and facilitate a more robust understanding of uncertainties in ML predictions. By enabling an extension of the GUM to include nominal properties, this framework not only enhances the reliability of machine learning …