Abstract
In 2016, the total value of fraud for credit cards issues within the SEPA region was €1.8 billion (Whatman, 2019). The company with which I am employed works in the domain of financial crime prevention software, so I chose to focus this project on one of the key challenges in this area. My project delivers an end-to-end solution that uses an Azure hosted predictive model, accessed via a separate Shiny R dashboard, to assess individual credit card transactions in real time for the likelihood of fraud.
A dataset of 25K+ historical US credit card transactions (2104), each one labelled as ‘Fraud’ or ‘Not Fraud’, is engineered to train and deploy a model to predict if ‘future’ transactions appear to be fraudulent.
The Machine Learning modelling process is managed through the online Microsoft Azure Machine Learning Studio (classic) platform. This includes the hosting of a REST Endpoint for the model, to be accessed as a Web Service by an external application for fraud prediction.
A separate Shiny R dashboard is included in this project to access the predictive fraud model through an API call, passing the details of ‘new’ card transactions as parameters one-by-one in real time to the Azure Web Service.
This Shiny dashboard is hosted on the ShinyIO platform and also provides a secondary interface to provide key data visualisation graphs on the original dataset.