Online

Machine Learning in Montpellier, Theory & Practice

I will mainly present the paper « Evaluating Probabilistic Classifiers: The Triptych » [ https://isdm.umontpellier.fr/doi.org/10.1016/j.ijforecast.2023.09.007 | https://isdm.umontpellier.fr/doi.org/10.1016/j.ijforecast.2023.09.007 ] that provides a trinity of evaluating plots for probability forecasts for binary events: Reliability diagrams, ROC curves and Murphy diagrams, which focus on the individual aspects of calibration, discrimination and overall predictive performance, respectively. Summary statistics for these properties are given by an associated score decomposition. I will furthermore present extensions of the underlying ideas such as statistical inference methods and extensions to point (e.g., mean or quantile) forecasts.

Paper Visio /Briefcase/RF-Inria_logo_signat.png" target="_blank" rel="noopener">Online