Multi-layered self-calibrating analytics

1. A fraud detection surveillance computer-implemented method for determining outlier events in a computing network, the method comprising:

in a fraud detection computing model comprising a plurality of self-calibrating models:

providing a first set of one or more variables to a first self-calibrating model;

providing a second set of one or more variables to a second self-calibrating model,

the one or more variables being generated from real-time transaction data the values of said one or more variables being associated with one or more transaction-related characteristic used to detect fraudulent transactions;

producing a first self-calibrating model output based on processing the first set of one or more variables according to the first self-calibrating model;

producing a second self-calibrating model output based on processing the second set of one or more variables according to the second self-calibrating model; and

combining the first self-calibrating model output with the second self-calibrating model output to generate an output characterizing whether a transaction associated with the transaction data is fraudulent,

tuning the first and second self-calibrating model outputs based on a first weight being assigned to the first self-calibrating model output and a second weight being assigned to the second self-calibrating model output,

the first weight and the second weight calculated and assigned in a non-uniform manner in real-time to respective outputs of the first self-calibrating model and the second self-calibrating model to more efficiently tune weights assigned in the fraud detection computing model,

the first weight and the second weight being different in value, at least one of the first weight or the second weight being assigned based on correlation analysis or full regression training, and

supervised training being used to enhance self-learning capabilities of the fraud detection computing model after the fraud detection computing model is in production, thereby requiring a limited amount of data to tune at least one of the first weight or the second weight.

2. The method in accordance with claim 1 ,

wherein the plurality of self-calibrating models are implemented in a hidden layer.

3. The method in accordance with claim 1 , wherein assigning a weight to a self-calibrating model output includes assigning a zero weight to a subset of one or more of the plurality of self-calibrating models to designate the subset as experimental self-calibrating model.

4. The method in accordance with claim 1 , wherein a linear combination of the one or more variables from a set of variables is used to determine a fraud score that is related to a number and a size of one or more outlier values associated with the one or more variables.

5. The method in accordance with claim 4 , wherein the one or more outlier values are based on a real-time computation of variable distributions of the one or more variables.

6. The method in accordance with claim 5 , wherein the one or more outlier values are updated recursively for each new set of transaction data.

7. A fraud detection computer-implemented method for detecting fraud in transaction data, the method comprising:

executing, by one or more data processors, a plurality of self-calibrating models based on values assigned to a plurality of variables, at least one of the one or more variables being generated from real-time production data related to the transaction data, at least one of the self-calibrating models being executed on a different set of one or more variables selected from the plurality of variables;

producing a self-calibrating model output for at least two of the plurality of self-calibrating models based on the selected one or more variables, at least one of the plurality of self-calibrating models using a linear combination of the one or more variables from a set of variables to determine a fraud score, according to an unconditional re-scaling across one or more independent variables based on:

q

⁡

(

x

i

❘

θ

)

≡

x

i

-

θ

i

,

1

θ

i

,

2

∈

[

0

,

C

]

,

where ((θ i,1 ,θ i,2 ∈θ) are location and scale parameters respectively of a computed distribution of independent variable x i , wherein the fraud score is related to a number and a size of one or more outlier values associated with the one or more variables;

combining values of the self-calibrating model outputs from the at least two of the self-calibrating models in an output model implemented by one or more data processors;

wherein the self-calibrating model outputs are tuned based on weights assigned to the self-calibrating model outputs,

the weights calculated and assigned in a non-uniform manner in real-time to more efficiently tune the weights assigned in the fraud detection computing model; and

generating an output for the real-time production data from the self-calibrating model output of the at least one of the self-calibrating models, the output representing a score relating to whether a transaction associated with the transaction data is fraudulent.

8. The method in accordance with claim 7 , wherein the one or more outlier values are based on a real-time computation of variable distributions of the one or more variables.

9. The method in accordance with claim 8 , wherein the one or more outlier values are updated recursively for each new set of transaction data.

10. The method in accordance with claim 7 , wherein assigning a weight includes assigning a zero weight to a subset of one or more of the plurality of self-calibrating models to designate the subset as experimental self-calibrating models.

11. The method in accordance with claim 7 , wherein combining the adjusted values of the self-calibrating models output includes combining a fraud score from at least two of the self-calibrating models to determine a score based on an inputted threshold.

12. A computer-implemented system comprising:

at least one programmable processor; and

a machine-readable medium storing instructions that, when executed by the at least one processor, cause the at least one programmable processor to perform operations comprising:

execute a plurality of self-calibrating models based on values assigned to one or more variables selected from a set of variables, at least one of the one or more variables being generated from real-time production data related to transaction data, at least one of the self-calibrating model being executed on a different set of one or more variables selected from the one or more variables;

produce a self-calibrating model output for at least two of the plurality of self-calibrating models based on the selected one or more variables, at least one of the plurality of self-calibrating models using a linear combination of the one or more variables from a set of variables to determine a fraud score related to a number and a size of one or more outlier values associated with the one or more variables;

combine the self-calibrating model outputs from the at least two of the plurality of self-calibrating models in an output model implemented by one or more data processors, wherein the self-calibrating model outputs are tuned based on weights assigned to at least two of the self-calibrating model outputs,

the weights calculated and assigned in a non-uniform manner in real-time to more efficiently tune the weights; and

generate an output for the real-time production data from the self-calibrating model outputs, the output representing a score relating to whether a transaction characterizing at least a portion of the transaction data is fraudulent.

13. The system in accordance with claim 12 , wherein the selection of one or more selected variables is based on factor group analysis to minimize correlation among selected variables for each of the plurality self-calibrating models.

14. The system in accordance with claim 12 , wherein the one or more outlier values are based on a real-time computation of variable distributions of the one or more variables.

15. The system in accordance with claim 14 , wherein the one or more outlier values are updated recursively for each new set of transaction data.

16. The system in accordance with claim 15 , further comprising assigning a weight to at least one self-calibrating model output.

17. The system in accordance with claim 16 , wherein the output model processes the output from at least one of the plurality of self-calibrating models according to the weight.

18. The system in accordance with claim 17 , wherein assigning a weight to at least one of the self-calibrating model outputs includes assigning a zero weight to a subset of one or more of the plurality of self-calibrating models to designate the subset as experimental self-calibrating models.