Grassroots Innovation Database [GRID] | A data science approach to control tax evasion in commercial taxation system

About the Practice

Problem Statement:

Fraudulent activity, unfortunately, is inherent in our society from time immemorial. It is primarily motivated by the unscrupulous desire of people to get personal benefits by exploiting the loopholes in the existing laws in a system. Certain types of fraudulent activities are easier to identify and scrutinize. On the other hand, there are fraudulent methods that are extremely difficult to track down due to the complexity of the processes involved in handling them. Tax evasion is the illegal evasion of taxes by business dealers, using thoughtful and well considered techniques. In this work, we focus on tax evasion occurring at both ends of the spectrum. For the same, we used the commercial tax dataset provided by the Government of Telangana, India. In the most rudimentary form of tax-evasion, the business dealer/entity do not file its tax return statements, we call them as return defaulters, and it is the most widely rampant form of tax-evasion. We focus on preventing this type of tax-evasion before it happens. In the other extreme, there is circular trading which is performed by few highly sophisticated business dealers and this evasion is in hundreds of crores of rupees. We developed algorithms to detect circular trading and reported our findings to the concerned authorities who can use this information for further scrutiny. The prior identification of potential tax evaders can be very useful to the taxation department as they can take proactive measures like sending alert messages and mails, and if necessary visit the business premises of potential tax defaulters to force them to file their future tax-returns. This kind of personalized scrutiny has led the tax evaders to avoid doing malpractices at least for a certain period of time. Our work towards controlling circular trading follows a postmortem approach. Here we analyze a huge amount of data (in the order of few terabytes) pertaining to a set of colluding dealers and devised customized techniques to re-engineer the fabricated social network. This helps in bringing out the real network connections which were hidden among the nexus of illegitimate transactions created by the colluding dealers. The problem of isolating the real transactions from the illegitimate ones is of significance to the taxation authorities as it makes the process of scrutinizing the colluding dealers very much easier for them.

Detail:

Tax is a mandatory financial charge or some other type of levy imposed upon an individual or a legal entity by a state in order to fund various public expenditures. There are mainly two type of tax structures, direct tax and indirect tax. Here we focus on the indirect tax structure. Indirect tax (e.g., Value added Tax (VAT) and Goods and Services Tax (GST)) is a tax collected by an intermediary (such as a retail store) from the person who bears the ultimate economic burden of the tax (such as the consumer). In this project, we collaborated with the Commercial Tax Department of Telangana, India, who shared their domain expertise and the recent six years' tax-returns dataset to develop techniques for handling tax-evasion occurring at various complexity levels. In the following paragraphs, we give an overview of this work. Circular trading is one of the major malpractices used in VAT evasion. Motivation for circular trading is to hide suspicious sales or(and) purchase transactions, which are otherwise easily detectable by taxation authorities. Dealers involved in circular trading collude with each other to hide their suspicious transactions by superimposing them by several illegitimate sales transactions. These illegitimate sales transactions are created within a short duration of time and it goes around in a circular manner without any value-add. Since there is no value-add for the illegitimate transactions, they do not pay VAT on these transactions and confuse the tax authorities about their suspicious transactions. In the first phase, we addressed the problem of circular trading by developing an algorithm that detects circular trading and removes the illegitimate cycles to uncover the suspicious transactions. We formulated the problem as finding and then deleting specific type of cycles in a directed edge-labeled multigraph. Upon running the algorithm on the tax-dataset we uncovered several suspicious transactions that were hidden among huge amounts of illegitimate transactions with no significant value addition. In the second phase, we developed a clustering algorithm that is customized to detect the group of colluding dealers who perform heavy illegitimate trading among themselves. The results obtained from running this algorithm contains groups of several colluding dealers that are of interest to the tax-enforcement officers. These two works are published in IEEE International Conference on Big Data Analytics 2018 (ICBDA 2018, Shanghai, China) and Computing Conference 2018 (London, United Kingdom), respectively. In GST, any business entity is required by the law to file their tax return statements following a periodical schedule. Avoiding to file the tax return statement is one among the most rudimentary forms of tax evasion. Dealers committing tax-evasion in such a way are called return defaulters. In the third phase of this work, we constructed a binary logistic regression model that predicts with 86% accuracy whether a business entity is a potential return defaulter for the upcoming tax-filing period. This is a valuable piece of information for the taxation authorities as they can take proactive measures, like sending alert messages, to potential return defaulters that may force them to file their tax returns. This work is published in IEEE International Conference on Data Science and Advanced Analytics (IEEE DSAA 2018, Turin, Italy). In the fourth phase of this project, we devised a technique to predict the amount of tax-revenue lost by the state of Telangana due to the unscrupulous actions from a particular set of business dealers. For the same, we built a linear regression model using the tax-return information of genuine business dealers and predicted the amount of tax evaded by suspicious business dealers. This work is accepted in IEEE/WIC/ACM International Conference on Web Intelligence (IEEE WIC, Santiago, Chile).

A Data Science Approach To Control Tax Evasion In Commercial Taxation System

About the Practice

About the Innovator

Practice Details