In addition to WAF, an additional way to protect an application is to prevent users from committing fraud. To do this, many use machine learning models.
Amazon Fraud Detector differs from traditional models (such as AutoML) in that it is ad hoc, the service performs data enrichment with Amazon and third-party sources, this means that the service learns from attempts to defraud Amazon.com and this also contribute to the intelligence that the service provides them (for example, through IP reputation).
Fraud Detector will facilitate the construction of the Machine Learning models, based on your historical data for training, and will produce a score per transaction
The service has a trial period of 60 days (free trial with limits, see pricing page)
The service has a page to check current usage and estimate future usage.