Our goal is to demonstrate how Machine Learning could help us to identify fraudulent transactions on web purchases. How to predict if a purchase is a fraudulent one of come from a regular customer ? We start with a dataset of 151.112 transactions in a classic .csv format (as it is in most banks)
Our task will be to
-
Identify country of origin for every user (from his/her IP adress)
-
Build a model able to predict if a purchase is fraudulent or not (with analysis of false positive/negative)
-
Explain the model to a non-ML sprecialiset stakeholder (e.g. the Head of Risk). Explain it from a user point of view