Network Risk Panel Follow

The Network Risk panel is derived from the Ekata Identity Network, our dataset of billions of online transactions used to detect legitimate and anomalous consumer behavior. Network Risk uses machine learning predictions to surface: 

• A score of the identity's risk behavior level – high, medium, or low
• 4 behavior categories – activity patterns of identity elements 
• 12 risk signals – indicating positive or negative behavior across that identity’s transaction history within our Network

About the Score

The Network Risk Score is trained on a random forest machine learning model, and as with every model, it must be trained on and fed with the right data. Our product and engineering teams tested hundreds of attributes that are within the Identity Network. After this testing, a combination of 147 elements were chosen based on their proven predictive value in helping customers separate fraudulent behavior from legitimate behavior. Our model then looks across these 147 features, giving equal weight to each input type, and ultimately surfaces a score and the top 12 features that influenced the score. These 12 features are the dynamic risk signals in the panel, i.e. primary address + email first seen together 1 year ago. 

Behavior Categories and their Risk Signals 

The 12 signals reveal key behavioral fraud and activity patterns of a consumer:

1. Velocity indicates the number of times an identity attribute was seen in recent transactions. 
1. Example - this phone number has been seen in X transactions over the last 90 days. 
2. Trend - once a fraudster steals an identity, they may try to commit fraud through multiple transactions at the same time before the identity is flagged.
2. Volatility indicates the number of times an attribute of an identity has changed over the last 90 days. 
1. Example - X phone numbers have been associated with this shipping address in the last 30 days. 
2. Trend - once a fraudster steals an identity, they may use different combinations of identity attributes to commit fraud. 
3. Popularity indicates the number of merchants where an attribute is seen in recent transactions.
1. Example - this phone number has been seen by X number of different merchants in the last 30 days.
2. Trend - once a fraudster steals an identity, they may try to commit fraud with multiple merchants at the same time before the identity is flagged.
4. Age measures the reliability of an identity element based on its history in Ekata's network
1. Example - this address and phone number were first seen together 1 day ago 
2. Trend - fraudsters typically have a shorter, more fragmented history