2021/02/12 by Kjersti Aas, Thomas Nagler, Aas, Kjersti +5 · 1 citation
Computer Science · Mathematics · #Data Analysis with R #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Methodology (stat.ME) #Statistical Methods and Inference #Statistical and Computational Modeling
paper · pdf · doi:10.48550/arxiv.2102.06416
openalex publication_date 2021/02/12 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
The original development of Shapley values for prediction explanation relied\non the assumption that the features being described were independent. If the\nfeatures in reality are dependent this may lead to incorrect explanations.\nHence, there have recently been attempts of appropriately modelling/estimating\nthe dependence between the features. Although the proposed methods clearly\noutperform the traditional approach assuming independence, they have their\nweaknesses. In this paper we propose two new approaches for modelling the\ndependence between the features.\n Both approaches are based on vine copulas, which are flexible tools for\nmodelling multivariate non-Gaussian distributions able to characterise a wide\nrange of complex dependencies.\n The performance of the proposed methods is evaluated on simulated data sets\nand a real data set. The experiments demonstrate that the vine copula\napproaches give more accurate approximations to the true Shapley values than\nits competitors.\n