ML Models & Configuration
Configuring Verisoul’s machine learning models for your business
Verisoul’s machine learning models are the core of our fraud prevention system. This guide explains how configuration changes directly affect the risk scores returned by our API endpoints.
Understanding Model Configuration
Every customer starts with the same default configuration optimized for general fraud detection. When you make API calls to Verisoul’s endpoints, the scores returned are calculated using your specific configuration. The Verisoul team can help tune these configurations to adjust how strict or lenient different signals are scored.
For example, configuration changes can:
- Lower the proxy risk score when a user is on Apple Private Relay
- Increase the risk score when a device is linked to multiple accounts
- Adjust how aggressively the API flags VPN usage
- Change scoring thresholds for bot detection
Available Configuration Options
The following parameters can be configured to adjust your API response scores:
Network Signals
- Proxy detection sensitivity
- VPN detection thresholds
- Datacenter IP scoring
- Tor exit node handling
- Impossible travel detection
Device Signals
- Device fingerprinting strictness
- Cross-device linking thresholds
- Bot detection sensitivity
- Behavioral biometrics weighting
Account Signals
- Multi-account detection thresholds
- Account linking sensitivity
- Email risk scoring for disposable/undeliverable domains
- Historical fraud patterns
Configuration Process
Configuration changes are managed by the Verisoul team to ensure optimal performance. To request configuration updates:
- Contact your Verisoul account representative
- Describe your specific needs and use cases
- Our team will analyze your requirements
- We’ll implement and test the changes
- Monitor the results and adjust as needed
Best Practices
When considering configuration changes:
- Start with default settings and monitor results
- Identify specific patterns of false positives/negatives
- Request targeted adjustments rather than broad changes
- Allow time to measure the impact of changes
- Consider your industry’s unique fraud patterns
Next Steps
- Explore Lists and Rules for additional fraud prevention
- Learn about Blocking Bad Actors with verifications
- See the Dashboard documentation for monitoring model performance