Machine Learning Rules

Updated on 09.02.22
1 minute to read
Copy link
Video Guide
A quick video guide explaining the Machine Learning Rules features of the SEON Sense Platform

Training the machine learning system by labelling manually reviewed transactions.

Machine Learning Rules

Automatically generated rules that are based on previous transactions and the feedback from labels and manual reviews can be seen under the Machine Learning Rules tab.

The SEON Machine Learning System can generate two main types of rules: 

  • Complex, multiple parameter rules
  • Heuristic rules with single parameters (e.g. IP = XYZ) which are triggered after the 2nd case. 

Each listed rule shows information on the accuracy as well as statistics on the number and value of transactions affected.

You may adjust accuracy thresholds to tighten or loosen triggering conditions.

Machine learning rules can be found under the fourth tab in the Scoring engine. Heuristic rules are listed here too.

 

Activating Machine Learning Rules

Any rule on the machine learning listing can be activated by selecting the switch next to the rule or by setting a minimum accuracy within the SEON Settings - in which case SEON will automatically enable any matching rules.


 

Selecting a machine learning rule in the list also allows for the specific rule parameters to be reviewed. It also allows for the rule to be converted to a custom rule - allowing for specific modifications to be made if required. In addition, It is also possible from this screen to review transactions flagged and test the rule on recent transactional data using the Rule Tester functionality.
 

Feedback

Based on feedback provided, the SEON Machine Learning module generates rule suggestions based on the last 180 days worth of data. These rules are visible in the Workbench for 7 days as well as in the Scoring Engine.

There are two main ways to provide feedback about and label transactions:

1. Marking the right state and label on the UI

2. Using the Label API. For example, using the postback data from payment gateway: authorized, lost or stolen etc.

On providing feedback, every transaction’s state should be set to one of the three main categories:

 

StateCategory
APPROVESafe transaction.
REVIEWSuspicious transaction, not confirmed fraud yet.
DECLINEConfirmed fraudulent transaction.

It is also possible to create sub-categories for each fraud status type allowing for the specific reasons for the categorisation of the transaction - for example “chargeback”, “bonus abuser” or “payment gateway”. This is possible on the Settings -> Machine Learning section. 

These can then be selected either when manually re-classifying transactions or when transactions are updated via the Label API. Additionally, you can set the ML algorithm to only create rules based on your feedback labels. 

Was this article helpful?

?Got a question

Talk to sales