Nota conforme: sistema integrado para classificação automatizada de serviços em NFS-E com machine learning

Abstract

This study presents the development of Nota Conforme, an integrated system based on Artificial Intelligence (AI) that employs Natural Language Processing (NLP) and Machine Learning (ML) techniques for the automatic classification of services described in the Electronic Service Invoice (NFS-e). The proposal arises from the need to optimize tax audits and address tax evasion resulting from the misreporting of services under incorrect tax rates. The methodology involved the collection and labeling of NFS-e data, text preprocessing, vector representation using TF-IDF, and the comparative evaluation of seven classification algorithms, which led to the selection of Random Forest as the final model. To enable practical application, the model was encapsulated in an API, allowing integration with existing tax systems. The results demonstrated high classification accuracy, highlighting the potential of the solution to assist auditors in identifying inconsistencies and enhancing tax inspection processes. The study concludes that the application of AI to the classification of services described in NFS-e represents a promising approach to improving revenue collection efficiency and mitigating tax losses.