Self-learning online multi-layer method for unsupervised risk assessment
A method of protecting a computer system from fraudulent use includes collecting and aggregating sets of risk predictor values for user-initiated events into user-specific aggregations and organization-wide aggregations, and in response to a current event initiated by a user, generating a risk indicator as a combination of a user-specific indicator and an organization-wide indicator based on current event parameters and the user-specific and organization-wide aggregations. Based on the risk indicator indicating that the current event may be a fraudulent use, a protective control action is taken (such as denying or modifying a requested access) to protect the computer system.
1. A method of operating a risk-engine computer, comprising:
loading, by the risk-engine computer, non-zero initial counts into a risk engine operated by the risk-engine computer, the initial counts indicating expected counts that risk model will acquire over time, the risk model, by the risk-engine computer, generating risk scores based at least in part on aggregate counts of (i) risk predictor values in respective categories for respective users, (ii) risk predictor values for respective risk predictors, and (iii) risk predictor values in respective categories;
executing, by the risk-engine computer in response to requests received from a protected computer system in communication with the risk-engine computer, the risk model over successive time windows, including (i) aging out the initial counts over a course of multiple successive time windows and (ii) acquiring new counts based on transactions processed by the risk engine after loading the initial counts; and
continuing to age out the initial counts while also aging out the new counts over further successive time windows, thereby enabling the risk model to adapt to changes over time,
wherein the method further comprises obtaining the initial counts based on a preliminary study of logins over a designated period of time, and, when acquiring new counts, selectively updating the counters to reflect the new counts based on whether authentication results produced by the risk engine indicate successful authentication.
2. The method of claim 1 , further comprising obtaining the initial counts based on counts accumulated from operation of a different instance of the risk model.
3. A computer program product including a set of non-transitory, computer-readable media having instructions which, when executed by control circuitry of a risk-engine computer, perform steps of operating the risk-engine computer, the steps comprising:
loading, by the risk-engine computer, non-zero initial counts into a risk engine operated by the risk-engine computer, the initial counts indicating expected counts that risk model will acquire over time, the risk model, by the risk-engine computer, generating risk scores based at least in part on aggregate counts of (i) risk predictor values in respective categories for respective users, (ii) risk predictor values for respective risk predictors, and (iii) risk predictor values in respective categories;
executing, by the risk-engine computer in response to requests received from a protected computer system in communication with the risk-engine computer, the risk model over successive time windows, including (i) aging out the initial counts over a course of multiple successive time windows and (ii) acquiring new counts based on transactions processed by the risk engine after loading the initial counts; and
continuing to age out the initial counts while also aging out the new counts over further successive time windows, thereby enabling the risk model to adapt to changes over time,
wherein the steps further comprise obtaining the initial counts based on a preliminary study of logins over a designated period of time, and, when acquiring new counts, selectively updating the counters to reflect the new counts based on whether authentication results produced by the risk engine indicate successful authentication.
4. The computer program product of claim 3 , where the steps further comprise obtaining the initial counts based on counts accumulated from operation of a different instance of the risk model.
5. A risk-engine computer, comprising control circuitry that includes a set of processing units coupled to memory, the control circuitry constructed and arranged to:
load non-zero initial counts into a risk engine operated by the risk-engine computer, the initial counts indicating expected counts that risk model will acquire over time, the risk model, by the risk-engine computer, generating risk scores based at least in part on aggregate counts of (i) risk predictor values in respective categories for respective users, (ii) risk predictor values for respective risk predictors, and (iii) risk predictor values in respective categories;
execute, in response to requests received from a protected computer system in communication with the risk-engine computer, the risk model over successive time windows, including (i) aging out the initial counts over a course of multiple successive time windows and (ii) acquiring new counts based on transactions processed by the risk engine after loading the initial counts; and
continue to age out the initial counts while also aging out the new counts over further successive time windows, thereby enabling the risk model to adapt to changes over time,
wherein the control circuitry is further constructed and arranged to obtain the initial counts based on a preliminary study of logins over a designated period of time, and, when new counts are acquired, selectively update the counters to reflect the new counts based on whether authentication results produced by the risk engine indicate successful authentication.
6. The risk-engine computer of claim 5 , where the control circuitry is further constructed and arranged to obtain the initial counts based on counts accumulated from operation of a different instance of the risk model.