Publication
Title
Shapley value for tax audit data valuation
Author
Abstract
Tax authorities have access to unprecedented levels of information on taxpayers. However, these vast amounts of data lead to new challenges, with tax authorities running the risk of being overwhelmed by the enormous amount of incoming data. We present a valuation technique to quantify the value of features for tax audit prediction. Our approach, rooted in the game-theoretical Shapley value, effectively assigns importance to features derived from various Directives on Administrative Cooperation within the European Union and the OECD's automatic exchange of information. We show that our results can be used for global explanations of the predictive model, feature selection and determining which data should be acquired or cleaned with priority, similar to active feature acquisition. Our results can assist tax authorities in managing the large amounts of data they receive under the different disclosure regulations.
Language
English
Publication
Antwerp : University of Antwerp , 2023
Volume/pages
24 p.
Full text (open access)
UAntwerpen
Faculty/Department
Research group
Publication type
Subject
Affiliation
Publications with a UAntwerp address
External links
Record
Identifier
Creation 26.09.2023
Last edited 29.09.2023
To cite this reference