Font Size: 

Shapley values are a practical tool from Explainable AI used to interpret model outcomes on the observation level. Their usefulness has also been demonstrated in the context of multivariate outlier detection, where the contributions of single variables to the overall outlyingness are evaluated. This allows for an alternative view to cellwise outlyingness, where the interest is in identifying deviating cells of a data matrix. The concept of outlier explanation based on Shapley values can be extended to outlyingness for matrix-valued observations, which is an interesting new topic in robustness by itself.