Research Article
Comparison of Algorithms for Classification of Financial Intelligence Reports
@INPROCEEDINGS{10.1007/978-3-031-22324-2_16, author={Roberto Zaina and Douglas Dyllon Jeronimo de Macedo and Mois\^{e}s Lima Dutra and Vinicius Faria Culmant Ramos and Gustavo Medeiros de Araujo}, title={Comparison of Algorithms for Classification of Financial Intelligence Reports}, proceedings={Data and Information in Online Environments. Third EAI International Conference, DIONE 2022, Virtual Event, July 28-29, 2022, Proceedings}, proceedings_a={DIONE}, year={2022}, month={12}, keywords={Financial intelligence report Artificial intelligence Machine learning}, doi={10.1007/978-3-031-22324-2_16} }- Roberto Zaina
Douglas Dyllon Jeronimo de Macedo
Moisés Lima Dutra
Vinicius Faria Culmant Ramos
Gustavo Medeiros de Araujo
Year: 2022
Comparison of Algorithms for Classification of Financial Intelligence Reports
DIONE
Springer
DOI: 10.1007/978-3-031-22324-2_16
Abstract
This work shows the application of machine learning algorithms to decide whether Financial Intelligence Reports analyzed by the Brazilian Federal Police should be investigated or archived. We explain how the suspicious financial operations found in the Financial Intelligence Reports are a crucial piece of information to combat money laundering crimes. We depict the processing of such reports, which are often used to initiate a police investigation. When that is not the case, the reports are archived. In this work, we propose using machine learning to analyze these reports. We trained and used three classification algorithms: Decision Tree, Random Forest, and KNN. The results show that most reports should be archived. While Decision Tree and Random Forest indicated that about 2/3 of the reports should be archived, KNN indicated that about 4/5 of them should be archived. In the end, this work shows the feasibility of automating the analysis of the Financial Intelligence Reports by the Brazilian Federal Police, despite the need for more adjustments and tests to improve the accuracy and precision of the models developed.
