Fight Fraud With Machine Learning
September 4, 2020 by Tamas Kadar
Fraudsters constantly adapt their game plan based on your line of defense. This is why your risk rules need to be precise and dynamic. And Machine Learning is the perfect tool to improve the precision of your risk scores in the long run.
Get started in seconds
Unsure where to begin? Don’t worry. SEON’s risk scoring engine comes with preset rules tailored to your industry. Easily switch on and off to see how they impact your fraud rates in real time.
– Preset rules based on vertical
– Custom rules
– ML-suggested rules
Full customization control
Instantly create and test out new risk rules. You can even ask SEON support to help create them, and witness how efficient the new rules are based on your own historical data. You also get to add your own fraud reasons and categories for better analysis and reporting.
Whitebox machine learning rules
Let SEON automatically suggest rules for you, in a clear, human-readable format based on the C5.0 algorithm. You remain fully in control of the suggestions, which come with an accuracy percentage based on rates of historical true positives.
Controlled automation
Set your own thresholds for when a ML-suggested kicks into action based on an accuracy percentage. This gives you full power to automatically improve your ruleset, or to test the new rules in a sandbox environment before deploying them.
Combine AI with human insights
Don’t let false positives and false negatives impact your results. SEON Sense gives you full power to provide fraud patterns via the GUI or the Label API, so the ML-suggested rules are increasingly tailored to your specific business model.
Test and measure complex rulesets
Implement the most complex rules with single parameters or several (heuristic vs complex), and test their efficiency on your existing data to calculate true positive and false positive values.
A sample model of the decision tree. At each split, the ratio of decline transactions is increasing on the right and decreasing on the left.
Fix data flaws
SEON automatically performs a string similarity analysis using the Jaro-Winkler, N-Gram, and Levenstein distance algorithms. This helps flag minor differences in email handles, user names or addresses, for instance, so you can be certain of your customer data accuracy.
FAQ – Frequently Asked Questions
Which Machine Learning algorithm does SEON USE?
SEON leverages the C5.0 algorithm, which is quickly becoming the industry standard for producing interpretable models via decision trees.
How fast is the Machine Learning fraud scoring algorithm?
The algorithm retrains itself every hour based on your business’s historical data and real-time data.
How is accuracy measured?
Each risk rule triggers a suggestion to REVIEW or DECLINE user action. The ratio between approved and declined states gives us a measure of how successful the rule is. For instance, a rule that declines 95 transactions and approves 2 of them, will have an accuracy score of 97.94%.
What if I don’t want Machine Learning suggestions?
You can easily enable or disable rules in your dashboard. SEON also lets you manually set the accuracy thresholds before a ML rule is automatically implemented.
Speak with a fraud fighter.
Click here
Tamas Kadar
Tamás Kádár is the Chief Executive Officer and co-founder of SEON. His mission to create a fraud-free world began after he founded the CEE’s first crypto exchange in 2017 and found it under constant attack. The solution he built now reduces fraud for 5,000+ companies worldwide, including global leaders such as KLM, Avis, and Patreon. In his spare time, he’s devouring data visualizations and injuring himself while doing basic DIY around his London pad.
SEON Resources
Sign up for our newsletter
The top stories of the month delivered straight to your inbox