Explaining predictive models using Shapley values and non-parametric vine copulas

In this paper the goal is to explain predictions from complex machine learning models. One method that has become very popular during the last few years is Shapley values. The original development of Shapley values for prediction explanation relied on the assumption that the features being described...

Descripción completa

Guardado en:

Detalles Bibliográficos
Autores principales:	Aas Kjersti, Nagler Thomas, Jullum Martin, Løland Anders
Formato:	article
Lenguaje:	EN
Publicado:	De Gruyter 2021
Materias:	prediction explanation shapley values conditional distribution vine copulas non-parametric 62g05 62h05 68t01 91a12 Science (General) Q1-390 Mathematics QA1-939
Acceso en línea:	https://doaj.org/article/225427ea4f4d4ef4bd46be33a51c8970
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

id	oai:doaj.org-article:225427ea4f4d4ef4bd46be33a51c8970
record_format	dspace
spelling	oai:doaj.org-article:225427ea4f4d4ef4bd46be33a51c89702021-12-05T14:10:46ZExplaining predictive models using Shapley values and non-parametric vine copulas2300-229810.1515/demo-2021-0103https://doaj.org/article/225427ea4f4d4ef4bd46be33a51c89702021-06-01T00:00:00Zhttps://doi.org/10.1515/demo-2021-0103https://doaj.org/toc/2300-2298In this paper the goal is to explain predictions from complex machine learning models. One method that has become very popular during the last few years is Shapley values. The original development of Shapley values for prediction explanation relied on the assumption that the features being described were independent. If the features in reality are dependent this may lead to incorrect explanations. Hence, there have recently been attempts of appropriately modelling/estimating the dependence between the features. Although the previously proposed methods clearly outperform the traditional approach assuming independence, they have their weaknesses. In this paper we propose two new approaches for modelling the dependence between the features. Both approaches are based on vine copulas, which are flexible tools for modelling multivariate non-Gaussian distributions able to characterise a wide range of complex dependencies. The performance of the proposed methods is evaluated on simulated data sets and a real data set. The experiments demonstrate that the vine copula approaches give more accurate approximations to the true Shapley values than their competitors.Aas KjerstiNagler ThomasJullum MartinLøland AndersDe Gruyterarticleprediction explanationshapley valuesconditional distributionvine copulasnon-parametric62g0562h0568t0191a12Science (General)Q1-390MathematicsQA1-939ENDependence Modeling, Vol 9, Iss 1, Pp 62-81 (2021)
institution	DOAJ
collection	DOAJ
language	EN
topic	prediction explanation shapley values conditional distribution vine copulas non-parametric 62g05 62h05 68t01 91a12 Science (General) Q1-390 Mathematics QA1-939
spellingShingle	prediction explanation shapley values conditional distribution vine copulas non-parametric 62g05 62h05 68t01 91a12 Science (General) Q1-390 Mathematics QA1-939 Aas Kjersti Nagler Thomas Jullum Martin Løland Anders Explaining predictive models using Shapley values and non-parametric vine copulas
description	In this paper the goal is to explain predictions from complex machine learning models. One method that has become very popular during the last few years is Shapley values. The original development of Shapley values for prediction explanation relied on the assumption that the features being described were independent. If the features in reality are dependent this may lead to incorrect explanations. Hence, there have recently been attempts of appropriately modelling/estimating the dependence between the features. Although the previously proposed methods clearly outperform the traditional approach assuming independence, they have their weaknesses. In this paper we propose two new approaches for modelling the dependence between the features. Both approaches are based on vine copulas, which are flexible tools for modelling multivariate non-Gaussian distributions able to characterise a wide range of complex dependencies. The performance of the proposed methods is evaluated on simulated data sets and a real data set. The experiments demonstrate that the vine copula approaches give more accurate approximations to the true Shapley values than their competitors.
format	article
author	Aas Kjersti Nagler Thomas Jullum Martin Løland Anders
author_facet	Aas Kjersti Nagler Thomas Jullum Martin Løland Anders
author_sort	Aas Kjersti
title	Explaining predictive models using Shapley values and non-parametric vine copulas
title_short	Explaining predictive models using Shapley values and non-parametric vine copulas
title_full	Explaining predictive models using Shapley values and non-parametric vine copulas
title_fullStr	Explaining predictive models using Shapley values and non-parametric vine copulas
title_full_unstemmed	Explaining predictive models using Shapley values and non-parametric vine copulas
title_sort	explaining predictive models using shapley values and non-parametric vine copulas
publisher	De Gruyter
publishDate	2021
url	https://doaj.org/article/225427ea4f4d4ef4bd46be33a51c8970
work_keys_str_mv	AT aaskjersti explainingpredictivemodelsusingshapleyvaluesandnonparametricvinecopulas AT naglerthomas explainingpredictivemodelsusingshapleyvaluesandnonparametricvinecopulas AT jullummartin explainingpredictivemodelsusingshapleyvaluesandnonparametricvinecopulas AT lølandanders explainingpredictivemodelsusingshapleyvaluesandnonparametricvinecopulas
_version_	1718371708154413056

Explaining predictive models using Shapley values and non-parametric vine copulas

Ejemplares similares