A major global financial company was looking to detect fraudulent transactions in real time. They were looking to block these transactions or notify the customers immediately, thereby avoiding the heavy cost of dealing with fraudulent transactions.
Historic data was ingested and a fully non-linear, supervised machine learning model was build and deployed. The model detected fraud in real time. Incoming transactions were assigned a probability based on the likelihood they were fraudulent. The model also provides details on which variables were significant in predicting fraud. The dataflow is automatically recorded for regulatory compliance. The model is especially designed to deal with highly imbalanced datasets that are typical of this domain.