The Client, an Automated Clearing House (ACH), providing customers with a suite of payment processing services. A 3rd party fraud detection system has been in place making all the payment data visible to the providing company. Datrics helped build an in-house system that detects suspicious transactions hosted on-premises, so that data does not leave the client's infrastructure.
Existing solutions posed data leakage risks and potential regulatory risks leading to the requirement of an in-house solution. The performance of the new solution expected to be at least on par with the existing 3rd party tool. Development and maintenance costs should not surpass running costs spent on the existing solution.
Datrics not only simplifies the process of data preparation but also helps to build sophisticated machine learning models in weeks, not months. This includes slicing and dicing the datasets, choosing the optimal ML model, deploying the model, and using it in production via Datrics API.
In order to leverage the existing system results, an initial machine learning model has been trained and validated on the known examples of fraud. After the initial phase, the gained accuracy for the entire dataset was 0.84 (F-Score), which is more than sufficient according to the industry standards. Additionally, the anomaly detection approach was used to address the potentially unknown cases of fraud missed by the initial system.
Finally, the users' historical data was utilized to build a time series forecasting model that predicts the possible parameters of future transactions. If the upcoming transaction is out of the expected range, it should be considered suspicious and reviewed manually. As users' behavior might change over time, the time series model also brought a dynamic generation of the transaction limit.
With Datrics, ACH obtained an effective fraud detection system that showed better results than the old subscription-based service, being also more cost-effective in the long run.
.svg)
