US 12,174,964 B2
Dynamic account risk assessment from heterogeneous events
Hung Wei Tseng, Atlanta, GA (US); and Kailash Patil, Atlanta, GA (US)
Assigned to Pindrop Security, Inc., Atlanta, GA (US)
Filed by PINDROP SECURITY, INC., Atlanta, GA (US)
Filed on Jan. 27, 2021, as Appl. No. 17/159,748.
Claims priority of provisional application 62/969,954, filed on Feb. 4, 2020.
Prior Publication US 2021/0240837 A1, Aug. 5, 2021
Int. Cl. G06F 21/57 (2013.01); G06F 21/55 (2013.01)
CPC G06F 21/577 (2013.01) [G06F 21/552 (2013.01); G06F 2201/835 (2013.01); G06F 2201/86 (2013.01); G06F 2221/034 (2013.01)] 16 Claims
OG exemplary drawing
 
1. A computer-implemented method for account risk assessment, the method comprising:
training, by the computer, a machine-learning model for generating an account risk score by executing the machine-learning model using a plurality of training feature vectors of a plurality of training risk contributions for a plurality of training account features;
obtaining, by a computer, event data for a plurality of events to access a user account by one or more user devices via a plurality of channels, wherein the plurality of channels include at least one telephony channel and at least one computing channel, the event data for each event including a plurality of account features corresponding to a first set of risk contributions associated with communications from a user device via a channel of the plurality of channels;
extracting, by the computer, using the event data originated via the plurality channels from the one or more user devices, a second set of risk contributions as a feature vector representing the plurality of account features, the second set of risk contributions converted from the first set of risk contributions extracted for the plurality of channels, wherein a risk contribution of the second set of risk contributions was converted based on a forgetting factor;
generating, by the computer, the account risk score for the user account associated with the plurality of events via the plurality of channels based upon applying the machine-learning model to the feature vector representing the second set of risk contributions; and
utilizing the risk score for user authentication or fraud detection.