LaundroGraph: usando o aprendizado profundo para apoiar os esforços de combate à lavagem de dinheiro
Nos últimos anos, as técnicas de aprendizado profundo provaram ser altamente valiosas para lidar com inúmeras pesquisas e problemas do mundo real. Pesquisadores da Feedzai, uma empresa de ciência de dados financeiros com sede em Portugal...
![](data:image/jpeg;base64,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)
Visualização UMAP das incorporações de transações produzidas por LaundroGraph???????????? para 5 clientes amostrados aleatoriamente. No gráfico esquerdo, as transações de saída são representadas com um marcador de círculo e as transações de entrada são representadas com um marcador X. As cores representam os diferentes clientes. No gráfico à direita, as transações são coloridas de acordo com sua pontuação de anomalia, com cores mais escuras denotando uma pontuação de anomalia mais alta. Crédito: Cardoso, Saleiro & Bizarro.
Nos últimos anos, as técnicas de aprendizado profundo provaram ser altamente valiosas para lidar com inúmeras pesquisas e problemas do mundo real. Pesquisadores da Feedzai, uma empresa de ciência de dados financeiros com sede em Portugal, demonstraram o potencial do aprendizado profundo para a prevenção e detecção de atividades ilícitas de lavagem de dinheiro.
Num paper apresentado na 3rd ACM International Conference on AI in Finance , a equipa da Feedzai apresentou o LaundroGraph, um modelo auto-supervisionado que poderá simplificar o complicado processo de revisão de grandes quantidades de interações financeiras à procura de transações suspeitas ou trocas monetárias. Seu modelo é baseado em uma rede neural de grafo, uma rede neural artificial (ANN) projetada para processar autonomamente grandes quantidades de dados que podem ser representados como um gráfico.
“Querendo fortalecer nossa solução AML, e depois de identificar as principais dificuldades com o atual processo de revisão AML, pensamos em soluções para superar esses desafios usando IA”, disse Mario Cardoso, Research Data Scientist da Feedzai, ao TechXplore.
"O AML é particularmente desafiador devido à escassez de rótulos e ao fato de que o contexto em torno dos movimentos financeiros, especificamente as entidades com as quais interagimos e as propriedades de cada transação, são cruciais para informar as decisões. Com esses desafios em mente, procuramos criar uma máquina abordagem de aprendizagem que pode apoiar analistas humanos e facilitar a revisão de AML."
A revisão das interações financeiras em busca de atividades suspeitas pode ser uma tarefa muito tediosa e demorada para os analistas humanos. Cardoso e seus colegas decidiram simplificar bastante essa tarefa usando técnicas de aprendizado profundo , que são conhecidas por serem particularmente boas na análise de grandes quantidades de dados.
LaundroGraph, o modelo que eles criaram, pode codificar clientes bancários e transações financeiras, transformando-os em representações gráficas significativas. Essas representações podem orientar o trabalho dos analistas de combate à lavagem de dinheiro, destacando movimentações anômalas de dinheiro para clientes específicos sem que eles precisem examinar históricos inteiros de transações.
"LaundroGraph gera representações de comportamento densas e sensíveis ao contexto desacopladas de quaisquer rótulos específicos", explicou Cardoso. "Ele faz isso explorando as informações estruturais e de recursos de um gráfico por meio de uma tarefa de previsão de links entre clientes e transações. Definimos nosso gráfico como um gráfico bipartido de transações de clientes, que criamos usando os dados brutos de movimentos financeiros."
Os pesquisadores da Feedzai avaliaram seu modelo em uma série de testes, avaliando sua capacidade de prever transferências suspeitas em um conjunto de dados de transações do mundo real. Eles descobriram que seu poder de previsão era significativamente maior do que o de outros métodos de linha de base projetados para apoiar os esforços de combate à lavagem de dinheiro.
"Dado que não requer rótulos, o LaundroGraph é adequado para uma ampla variedade de aplicativos financeiros do mundo real que podem se beneficiar de dados estruturados em gráficos", disse Cardoso. "Nosso artigo propõe aproveitar essas incorporações para fornecer informações que podem acelerar o processo de revisão da detecção de AML, mas essa abordagem pode ser estendida a outros casos de uso (por exemplo, fraude) e as incorporações podem servir a uma ampla variedade de propósitos além das percepções analisamos (por exemplo, enriquecimento de recursos)."
No futuro, a LaundroGraph poderá auxiliar analistas financeiros e agentes anti-lavagem de dinheiro em todo o mundo na revisão de grandes quantidades de transações financeiras , ajudando-os a identificar atividades anômalas com mais rapidez e eficiência. Cardoso e seus colegas agora planejam desenvolver ainda mais seu modelo, ao mesmo tempo em que exploram seu potencial para resolver outros problemas financeiros.
"Direções futuras para nossa pesquisa incluirão experimentação em casos de uso adicionais, como fraude, e pesquisa em outras percepções/tarefas que podem ser ativadas ou aprimoradas por meio de incorporações, por exemplo, usando as incorporações como um ponto de partida informativo para um rótulo -escassas previsões a jusante", acrescentou Cardoso.
Mais informações: Mário Cardoso et al, LaundroGraph: Aprendizagem de representação gráfica auto-supervisionada para lavagem de dinheiro, 3ª Conferência Internacional da ACM sobre IA em Finanças (2022). DOI: 10.1145/3533271.3561727