A model can only be useful when it is put into operation. Typically, a data scientist or a team of data scientists will spend considerable time analyzing the raw data and build a machine learning model from it. Once a model is ready, it can be used on a web portal, workflow engine, or in a batch program to make predictions.
Our model is intended to detect anomalies in transaction data signaling suspicious transactions. Once we detect an anomalous transaction then we can take several measures to prevent the customer suffering a loss, such as putting a transaction on hold while waiting for the customer to confirm if it is valid by phone.
To put this model to use, we will create a data stream that simulates incoming real-time transactions. We should convert the incoming transactions into a form suitable for our fraud detection model created in the previous section. It means that the input to the model should be in the form 1, 3, 75, where 1 is the day...