Modelo de aprendizagem de máquina segmentado para previsão analítica de arrecadação fiscal baseada em informações de notas fiscais eletrônicas
In Brazil, the Tax on Operations Related to the Circulation of Goods and the Rendering of Interstate and Intermunicipal Transportation and Communication Services, known by the acronym ICMS, holds significant prominence in the revenue of the federative units, accounting for approximately 90%. As a co...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis de maestría |
| Estado: | Versión publicada |
| Fecha de publicación: | 2024 |
| País: | Brasil |
| Institución: | Universidade Federal do Maranhão (UFMA) |
| Repositorio: | Biblioteca Digital de Teses e Dissertações da UFMA |
| Idioma: | portugués |
| OAI Identifier: | oai:tede2:tede/5484 |
| Acceso en línea: | https://tedebc.ufma.br/jspui/handle/tede/5484 |
| Access Level: | acceso abierto |
| Palabra clave: | regressão; CNN; MLP; aprendizagem segmentada; previsão de receita fiscal; regression; segmented learning; tax revenue forecasting. Ciência da Computação |
| Sumario: | In Brazil, the Tax on Operations Related to the Circulation of Goods and the Rendering of Interstate and Intermunicipal Transportation and Communication Services, known by the acronym ICMS, holds significant prominence in the revenue of the federative units, accounting for approximately 90%. As a consumption tax, its value depends on economic activity, whose tax information the taxpayers record in electronic invoices issued to the tax agencies of each federative unit. This work proposes a learning model to predict ICMS revenue through a dataset derived from tax information in electronic invoices. The learning model uses a segmented approach that starts with splitting the training and validation datasets according to a given parameter. After that, the architecture fits several machine learning models for each split subset (segment). Finally, the architecture chooses the fit machine learning model (learning instance) that produces the best prediction result for each segment. These learning instances compose a hybrid instance set to predict the records of a test dataset. When comparing the results of the traditional approach without segmentation of the training and validation bases with those of the proposed model, an improvement in the metrics used was obtained by 29.52% and 18.4% for the years 2021 and 2022, respectively. And when comparing the results of the current model of the tax body that supported this research, the proposed model improved the metrics by 60.34% and 51.9% for the years 2021 and 2022, respectively. The improvement in the assertiveness metric used suggests that the model is promising in providing information to support decision-making in public management. Furthermore, the model can assist in solutions to combat tax evasion, analyze impacts resulting from changes in the tax system, and address other related challenges. |
|---|