Predictive Models In The Assessment of Tax Fraudevidences

The aim of the work is to verify the possibility of predictingthe result of future tax actions based on the results of the inspectionsalready carried out. The analysis of the information about the process,obtained from the auditors involved in the selection of taxpayers andin the inspection of companies, allowed the identification of the vari-ables used in the models, and the literature review allowed to identifythe techniques and tools necessary for their creation and training. Theresearch identified predictive models of logistic regression and neuralnetworks, whose predictions identified sets of companies that correspondto approximately half of the audited companies and account for morethan 80% of the constituted credit (89% in the case of the model neu-ral network), of so that these models have the potential to optimize theapplication of available resources and maximize results, assisting in theselection of indications of irregularities and fraud with greater potentialfor the constitution of the due credit.