Multivariate risk assessment via Poisson shelves
Detecting fraudulent activity can be a complex, manual process. In this paper, we adapt statistical properties of count data in a novel algorithm to uncover records exhibiting high risk for fraud. Our method identifies shelves, partitioning data under the counts using a Student's t-distribution. We apply this methodology on a univariate dataset including cumulative results from phone calls to a customer service center. Additionally, we extend this technique to multivariate data, illustrating that the same method is applicable to both univariate and multivariate data.
1. A system for performing risk assessment, comprising:
a memory comprising executable instructions;
a processor configured to execute the executable instructions and cause the system to:
access a probabilistic model in a library of known probabilistic models; modeling said count data using a Poisson distribution using the probabilistic model running on a processor;
calculate a threshold value, ω, using the probabilistic model;
partition the count data based on the threshold ω, using the probabilistic model;
determine a number of surviving records from among the partitioned count data based on the threshold ω, using the probabilistic model;
identify at least one shelf in the surviving records; and
define outliers for risk assessment according to the at least one shelf.
2. The system of claim 1 , wherein the executable instructions, when executed, cause the system to identify the at least one shelf by identifying an abrupt variation in generative parameters of a sequence of data.
3. The system of claim 1 , wherein the executable instructions, when executed, further cause the system to create a visual representation of the surviving records for identifying the at least one shelf.
4. The system of claim 3 , wherein the at least one shelf is user-identified is based on the visual representation.
5. The system of 4 , wherein the executable instructions, when executed, cause the system to identifying an abrupt variation in generative parameters of a sequence of data.
6. The system of claim 1 , wherein the at least one shelf is identified using concept drift detection.
7. The system of claim 1 , wherein the threshold value ω is calculated according to:
ω=λ+ n ×stdev( X )=λ+ n√λ,
where n is the number of standard deviation units in ω, λ is a mean of aPoisson distribution and X is a random variable having integer values.
8. The system of claim 7 , wherein the executable instructions, when executed, further cause the system to determine a number of datapoints greater than the threshold value ω for a given number of standard deviation units n.
9. The system of claim 1 , wherein the library of known probabilistic models is Python Pomegranate.
10. A non-transitory computer readable medium comprising instructions that, when executed by a processor of a processing system, cause the processing system to perform a method, the method comprising:
accessing a probabilistic model in a library of known probabilistic models; modeling said count data using a Poisson distribution using the probabilistic model running on a processor;
calculating a threshold value, ω, using the probabilistic model;
partitioning the count data based on the threshold ω, using the probabilistic model;
determining a number of surviving records from among the partitioned count data based on the threshold ω, using the probabilistic model;
identifying at least one shelf in the surviving records; and
defining outliers for risk assessment according to the at least one shelf.
11. The non-transitory computer readable medium of claim 10 , wherein identifying the at least one shelf comprises identifying an abrupt variation in generative parameters of a sequence of data.
12. The non-transitory computer readable medium of claim 10 , the method further comprising creating a visual representation of the surviving records for identifying the at least one shelf.
13. The non-transitory computer readable medium of claim 12 , wherein the at least one shelf is user-identified based on the visual representation.
14. The non-transitory computer readable medium of claim 13 , wherein identifying the at least one shelf comprises identifying an abrupt variation in generative parameters of a sequence of data.
15. The non-transitory computer readable medium of claim 10 , wherein the threshold value ω is calculated according to:
ω=λ+ n ×stdev( X )=λ+√λ,
where n is the number of standard deviation units in ω, λ is the mean of a Poisson distribution and X is a random variable having integer values.
16. The non-transitory computer readable medium of claim 15 , the method further comprising determining a number of datapoints greater than the threshold value ω for a given number of standard deviation units n.
17. The non-transitory computer readable medium of claim 10 , wherein the library of known probabilistic models is Python Pomegranate.