Demystifying Tax Evasion Using Variational Graph Autoencoders

Mehta, P. et al (2021) Demystifying Tax Evasion Using Variational Graph Autoencoders. In: 0th International Conference on Electronic Government and the Information Systems Perspective, EGOVIS 2021, 27 September 2021 through 30 September 2021, Virtual, Online.

Full text not available from this repository. (Request a copy)

Abstract

Indirect taxation is a significant source of income for any nation. Tax evasion hinders the progress of a nation. It causes a substantial loss to the revenue of a country. We design a model based on variational graph autoencoders and clustering to identify taxpayers who are evading indirect tax by providing false information in their tax returns. We derive six correlation parameters (features) and three ratio parameters from the data submitted by taxpayers in their returns. We derive four latent features from these nine features using variational graph autoencoder and cluster taxpayers using these four latent features. We identify taxpayers located at the boundary of each cluster by using kernel density estimation, which is further investigated to single out tax evaders. We applied our method to the iron and steel taxpayers data set provided by the Commercial Taxes Department, the government of Telangana, India.

[error in script]

IITH Creators: