Machine Learning Overview

SEON's machine learning-driven fraud and risk prevention system was designed to mitigate fraud and compliance-related losses, decrease operational costs, and facilitate scalability.

Machine learning algorithms can identify patterns and typical behaviors behind fraudulent transactions, which can help you catch them earlier and with improved accuracy. Our machine learning solution can process hundreds or even thousands of approved and declined transactions to discover connections you didn't even know existed. 

The machine learning models will create new human-readable rules (whitebox) or assign a fraud probability score to each transaction (blackbox). 

Both of these options are available in SEON, and you can decide how to use them.

Machine learning starts with learning from the right examples of what is good and bad in a business, continues through using the right algorithms, and lands the business impact with reliable, precise predictions ready to use. The following elements make these happen:

• Labels and transaction states serve as machine learning's data input. Adding verification labels to transactions builds the training data to create machine learning models on an ongoing basis.
  • Find out how it works the best here
• Machine Learning training and model evaluation is the processing of data using the applicable algorithms, producing and maintaining the best models in use.
  • Machine learning rules are only created when their accuracy exceeds 85%, evaluated on past transactions. This ensures that only relevant and powerful ML rules are created.
  • Blackbox Score’s unique model management system evaluates models daily. The best-performing model calculates scores. Daily trained models are evaluated against the current in-production model. The new model replaces the old one only if it performs better.
• SEON’s machine learning provides key input to decisions directly and indirectly.
  • To automate fraud prevention with machine learning the following ways: use Machine Learning rules, Email Cluster rules, and Blackbox score for decisions right away. 
  • Edit and adapt ML rules and email cluster rules to your needs, then enable their usage in the scoring engine to give you a shortcut to finding the best rules with the least effort.
  • Combine Blackbox score with rules to segment decisioning to high and low fraud likeliness transactions and improve fraud detection.

Our whitebox and blackbox solutions are a great addition to your team's fraud-fighting toolkit. Choose the one that best fits your needs, or use them in combination for maximum efficiency. The choice is yours.

Both can be automated and customized to your liking, so you can find a setup that's just right for you and your business.

* Whitebox machine learning and Email clustering requires account-specific training data to begin recommending rules: at least 1000 transactions with 100 in the DECLINE and 100 in the APPROVE state.

Whitebox Machine Learning

Our whitebox machine learning algorithms analyze your transaction history to discover fraud patterns and create rules suggestions. You can then implement its fully explainable rules to block certain user actions, such as suspicious logins, identity theft, or fraudulent transactions.  

Learn more: Stay in control with our Whitebox Machine Learning feature overview.

Email Clustering

Fraudsters use various email addresses, some complex, others simple, making it difficult to distinguish real ones from newly created mailboxes. AI-powered Email Clustering by SEON helps identify and group algorithmically generated email addresses, enhancing fraud detection.

Learn more: Explore Email Clustering feature overview.

Blackbox Machine Learning

Similar to the whitebox model, blackbox trains itself on historical data to help you find fraud patterns that stay hidden from the human eye. The difference lies in transparency — while you can see how the algorithm reached a conclusion in a whitebox model, the way blackbox comes to conclusions is more nuanced. The model can make surprising connections and discover behaviors that would be hard to translate to a human reader.  

SEON Advanced Digital Footprint's Machine Learning Network Scores use specialized blackbox models to instantly calculate email and phone network scores when you start using SEON. These models utilize sanitised, cross-customer data from SEON’s proprietary consortium dataset, providing immediate fraud risk assessment for email addresses and phone numbers.

Learn more: Dive into the details of our Blackbox score and Network Scores overview.

Feedback Loops

At its heart, machine learning algorithms learn from good and bad examples. Our machine learning solutions learn from transaction labels and take states resulting from rules or manual investigations into account. Using positive and negative transaction labels to give the most accurate feedback and rule out false positives. Providing feedback labels to SEON Machine Learning to optimize the performance and customize to the use case you want to apply Blackbox Score and Machine Learning Rules

Learn more: Read how to provide feedback manually or through the Label API.