What is Amazon Fraud Detector?
Amazon Fraud Detector is a fully managed service that helps you detect suspicious online activities such as the creation of fake accounts and online payment fraud without knowing the user. Amazon Fraud Detector uses machine learning (ML) to automatically identify any fraudulent activity, and report to the users, so you can catch more fraud, faster time With Amazon Fraud Detector, you can create a fraud detection ML model with just a few clicks and use it to evaluate online activities in milliseconds. And also you can deploy it on your production workloads, where your business activities include online payment safe.
How does the fraud detection mechanism work?
This mechanism involves various steps starting from the event you define to the Amazon Fraud Detection mechanism to implement the mechanism in your AWS applications through the Amazon Fraud Detection API.
In the first step, the user has to define the event where he suspects fraudulent activities may happen depending on the environment. Upload your historical events of such suspicious fraudulent activities to Amazon S3 and select a fraud detection model type.
Then on the second step, the Amazon Fraud Detection mechanism analyzes the uploaded historical data to form a custom model which automatically inspects and enriches your data, performs feature engineering, selects algorithms, trains and tunes your model, and hosts the model on your AWS resources.
Using the defined custom model, Amazon Fraud Detectors create rules to either accept, review or collect more information based on model predictions that analyses your data. Also, the Amazon Fraud detector creates application APIs to receive real-time fraud predictions and take action based on your configured detection rules.
In this blog, we are summarizing the steps that are involved in creating an event detection that occurs frequently in your environment and creating rules which deny any fraudulent transactions.
Amazon Fraud Detection Service available regions
Currently, Amazon Fraud Detection Service is available only in these regions. They are
- Europe (Ireland)
- Asia Pacific (Singapore)
- Asia Pacific (Sydney)
- US East (N. Virginia)
- US East (Ohio)
- US West (Oregon)
Choose any one of these regions on your AWS console and do the following steps explained.
Components of an Amazon Fraud Detector mechanism
Detectors – Detectors are comprised of models and rules that evaluate events for fraud.
Models – These are machine learning models library which can help increase the efficiency and accuracy of risk classification within your organization
Events – An event type defines the structure for an event sent to Amazon Fraud Detector. With Amazon Fraud Detector, you generate fraud predictions for events
Entities – An entity type classifies who is performing the event. During prediction, specify the entity type and id to define who performed the event.
Outcomes – An outcome is the result of a fraud prediction. Create an outcome for each possible fraud prediction result
Labels – A label classifies an event as fraudulent or legitimate. Labels are used to train supervised machine learning models.
Filters – Filters are used to search past predictions and select results to review the data received for the event, the detection logic applied during the prediction, and the conditions that resulted in a fraud prediction outcome
Defining an event to evaluate for fraud detection action
Login to AWS console using root user credentials. Choose the service Amazon Fraud Detector from All services → Machine learning → Amazon Fraud Detector. Click create an event to start with, as shown in the below screenshot.
On this event creation page, you need to create a name and a new entity. The entity represents who is performing the event. Example entities include customer, merchant, or account. In our example, we are creating an event called “vembudemoregistration” with the entity name “vembudemoentity”
In the event variables section, users need to define the event variable based on the collection of related variables. You can define the variable based on the selection of variables from your selection list or from a training data sheet, which you can upload from an existing datasheet.
You can define the variables that will be sent as a part of this event type. By adding them to the list below, you can later use these variables in models and rules. You can create a new variable based on the criteria like email address, IP address, user agent name, Payment card BIN number, Phone number, Billing Address, etc.
You can also provide labels and tags for this event but are optional. The below screenshot shows an example of creating an event by providing an event variable “Payment card Bin no” of a consumer to check the payment card’s genuinity and there is no fraudulent transaction.
Click create an event type
In this stage, the Amazon Fraud Detection mechanism asks you to build a machine learning model to deduct fraud on registration events. In our case, the payment card genuinity is checked before using the card on the registration phase itself.
Click Build a model to proceed further.
Defining & configuring a new model to build
In this step, We need to provide a model name and model type. Here the model type will be “Online Fraud Insights”. You can choose this model from the list-down menu as shown in the picture below. Also, you need to provide an IAM role to access the data sets you uploaded to the s3 location.
Before configuring this new model, it is advised to upload a CSV file that contains the historical data of fraudulent transaction details. The dataset must include the reserved headers EVENT_TIMESTAMP and EVENT_LABEL and at least two of the variables defined in the event type. After uploading the CSV file, the user must provide the data location along with the IAM role as shown in the below screenshot.
Click Next to configure your model, and click Next to review and recreate the model.`
Conclusion :
Amazon Fraud Detector is a very useful tool for fraud analysts. With just a few clicks, fraud analysts can enhance model detection with business rules that help control model behavior then deploy results as production-ready APIs to start generating predictions. Within a single console, you can build, deploy, and manage fraud detection models in minutes. You can also automate all the steps from data validation to model deployment and can be done with no machine learning or coding experience required.
Follow our Twitter and Facebook feeds for new releases, updates, insightful posts and more.
