XBRL Open Information Model for Risk Based Tax Audit using Machine Learning

Abstract

Tax audit is an effective instrument for preserving tax
compliance, and risk-based tax audit selection can optimize
it. Risk-based tax audit selection selectively auditing on high
financial risk wealthy taxpayers. In contrast, manually
selecting amid the plethora of taxpayer data is difficult,
prone to human error, costly and time-consuming.
Fortunately, using Extensible Business Report Language
(XBRL) as a well-known financial statement reporting
standard enables automation. This project proposed
software named XAFR as a model for extracting,
transforming, and loading the latest XBRL Open Information
Model (OIM) 1.0 standard US-SEC dataset and provided it as
a data source for risk classification using rule-based risk
scoring and Machine Learning. Several thorough testing
exposed Random Forest classifier as the best model for
Machine Learning risk classification with high accuracy,
revealing the excellent collaboration of rule-based risk
scoring approach with Machine Learning for risk
classification and the importance of XBRL as a transparent
but robust report standard that tax authorities can utilize.
The excellent system integration resulted in the ability to
expose wealthy high-risk taxpayers and high-risk industries
and predict risk classification based on two-year financial
statements. Moreover, this report introduces the critical
importance of RCA (Risk, Current Ratio, Assets) analysis and
SIC (Standard Industry Classification) utilization to generate
risk classification, rank and explanation. This project utilizes
financial indicators in the limited year and leaves the
semantic analysis for future works because of time and
hardware limitations. The possibility of predicting the
possible tax debt prediction are promising Machine Learning
future developments