Our deep learning based fraud detection technology is a cloud based implementation. The technology is integrated deeply with almost every other component of our solution and allows us to: 

1. PROTECT AGAINST PAYMENT CLEARING FRAUD FOR ACH TRANSFERS

2. DETECT AND AVOID TRANSACTION & PAYMENT FRAUDS

To achieve that we have a build a deep learning based model that is simply used to predict the risk of fraud. We are constantly tweaking and improving our implementation. Our scientists and engineers are dedicated to building state of the art fraud prevention system. 

Although we are experimenting with new parameters and training out neural network with different parameters, holdout data and training sets, we keep the stable version of our network separate from all experiments. We are dedicated to publishing the most up to date configurations on our website and will provide update schedules to merchants on an as needed basis.

The fraud detection implementation is based on multinomial logistic regression that places the risk of the fraud in 3 possible categories. The three categories are depicted in Figure 1.3

These categories are configurable by the merchants so they can decide how much risk they are comfortable with. If merchants choose to set the risk profile themselves then our deep learning based model only predicts the risk of fraud. However, if the merchants do not choose to set up the risk profile themselves, then our sophisticated adaptive fraud protection technology sets the risk thresholds dynamically. 

The benefit of letting the system decide the risk thresholds is that if the system observes an overall high risk of fraud in the system, it automatically switches to a risk averse profile by setting the fraud thresholds accordingly.