Pointmount Founder argues for behavioral biometrics accuracy measurement based on actions
The metrics generally used to measure the accuracy of biometrics are not as effective when applied to behavioral biometrics, Pointmount Inc. Founder Justin Macorin claims in a LinkedIn post.
False acceptance rate (FAR) and false rejection rate (FRR) measure biometric accuracy and performance, with lower numbers indicating a more accurate system, Macorin explains. For behavioral biometric data like keyboard, mouse and touch dynamics, however, producing these statistics without reference to the time and data volume measured will not reveal much about the accuracy of the system.
“We must understand how time and circumstance affect accuracy levels,” Macorin writes. “For example, a user may get up to get a coffee or go to the bathroom; during this time, behavioral analysis is impossible to perform as no data is actively generated.”
Keyboard dynamics rely on “keypress” events, and mouse and touch dynamics likewise rely on user activity. Because people may type differently depending on circumstances such as the time of day, even when activity is carried out to generate behavioral data, variation needs to be eliminated from the equation to get meaningful results.
Macorin has some suggestions for how to do so with each behavioral data type. The number of user actions being measured, as opposed to a duration of time, as would be considered for instance in voice biometrics, is a more useful basis from which to generate FAR and FRR.
A recent report from Juniper Research urges companies to adopt behavioral biometrics to head off increasing ecommerce fraud.