Percorrer por autor "Chostak, Christian"
A mostrar 1 - 1 de 1
Resultados por página
Opções de ordenação
- Machine Learning powered serverless fraud detectionPublication . Chostak, Christian; Costa, Ricardo André FernandesThere is an increasing concern about fraud in all market sectors. Although there is a great fuzz about fraud and fraud detection, just a small fraction of it was fully incorporated into real world applications. Counterfeited documents are reproductions or imitations of the originals ones. The present work aims to fulfill a gap in fraud analysis by automating and identifying those documents in seconds. Generally speaking, a payload containing a suspect fraudulent document will reach an Application Programming Interface gateway, which will redirect the request to Lambda functions and based on the event store it on SQS - Simple Queue Service, this queue will trigger a fleet of micro-services powered by Lambda functions as well. The non-exhaustive list of functions will proceed to read this queue and in the first moment create the metadata of the received document, registering on a Serverless Relational Database, whilst storing the document itself on S3 - Simple Storage Service. After that, it will call the second batch that will start the process of machine learning on the already saved image. Triggered by the finished process, a message will go to the SNS - Simple Notification Service - alerting the user. The output of the given analysis contains a sample of the input document showing where the fraud is if there is one. With the percentage and area given, the operator will be able to see what portion of the image was considered a fraud and from that moment forward, the user will have technical basis to accept the document or not.