TWO STAGE RISK MODEL BUILDING AND EVALUATION

• US 20150262184A1
• Filed: 03/12/2014
• Published: 09/17/2015
• Est. Priority Date: 03/12/2014
• Status: Abandoned Application

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Wang, Shoou-Jiun, Hobart, Justin, Zhang, Angang
• Application Number:
  US14/205,715
• Publication Number:
  US 20150262184A1
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06Q 10/067 Enterprise or organisation modelling
• CPC Class Codes: G06Q 20/4016 involving fraud or risk level assessment in transaction processing
• CPC Class Codes: G06Q 90/00 Systems or methods specially adapted for administrative, commercial, financial, managerial or supervisory purposes, not involving significant data processing

Abstract

A two stage model in which the first stage of the model applies a different weighting schema to different types of transactions in a transaction-based system is described. The first stage of the model focuses on capturing currently known patterns that indicate bad transactions. The second stage of the model focuses on rejecting transactions that are approved by the current model, with the objective of maximizing a measurable goal. Business knowledge is used to lower the cost of finding the optimal solution of the model by estimating parameters provided to the first and second stages of the model. Evaluation of the model accounts for retry transactions and for churn. Parameters associated with the model that maximizes the goal can be selected for a future model.

Claims

1. A system comprising:

at least one processor;

a memory connected to the at least one processor; and

at least one module that when loaded into the at least one processor causes the at least one processor to;

build a first stage of a two stage risk model, the first stage of the two stage risk model focused on capturing currently known patterns associated with bad transactions; and

build a second stage of the two stage risk model, the second stage rejecting transactions that are approved by a current risk model but that are actually bad.

2. The system of claim 1, further comprising a module that when loaded into the at least one processor causes the at least one processor to:

relabel transactions received from the current risk model so that rejected transactions received from a good user are relabeled “

good”

.

3. The system of claim 1, further comprising a module that when loaded into the at least one processor causes the at least one processor to:

relabel transactions received from the current risk model so that rejected transactions received from a bad user are relabeled “

bad”

.

4. The system of claim 1, further comprising:

a module that when loaded into the at least one processor causes the at least one processor to;

assign a first weight in a weighting schema to a transaction of a first type comprising an actually good transaction that has been approved by the current risk model;

assign a second weight in the weighting schema to a transaction of a second type comprising a transaction rejected by the current model, wherein a status associated with the transaction of the second type is unknown;

assign a third weight in the weighting schema to a transaction of a third type comprising an actually bad transaction approved by the current risk model.

5. The system of claim 4, further comprising:

a module that when loaded into the at least one processor causes the at least one processor to;

generate the first stage of the two stage risk model by performing a weighted logistical regression on a plurality of transactions, the plurality of transactions including relabeled transactions;

score the plurality of transactions;

discard a specified percentage of top-scored transactions;

discard a specified percentage of lowest-scored transactions;

discard transactions rejected by the current risk model; and

provide remaining transactions to the second stage of the two stage risk model, the second stage generated by performing an equal weighted logistic regression on the remaining transactions.

6. The system of claim 5, further comprising:

a module that when loaded into the at least one processor causes the at least one processor to;

evaluate the performance of the current risk model and the two stage risk model based on maximizing a measurable metric using a first parameter comprising a discount rate to account for multiple transactions placed by a good user whose transaction is rejected and based on a second parameter comprising a discount rate to account for multiple transactions placed by a bad user whose transaction is rejected.

7. The system of claim 1, further comprising:

a module that when loaded into the at least one processor causes the at least one processor to;

instantiate a plurality of instances of the two stage risk model, wherein each instance of the plurality of instances is provided with a different set of first and second parameters;

determine the instance maximizing revenue, the revenue for each instance measured by comparing total net profit produced by each instance; and

use the set of first and second parameters associated with the instance that maximized revenue for a future model.

8. A method comprising:

creating a two stage risk analysis model within the memory of a computer, the two stage risk analysis model comprising a first stage of risk analysis for a computerized transaction-based system and a second stage of risk analysis for the computerized transaction-based system, the first stage of risk analysis identifying patterns of characteristics of bad transactions, the second stage of risk analysis rejecting actually bad transactions approved by a current risk model.

9. The method of claim 8, further comprising:

relabeling at least one of a plurality of transactions received from the current risk model by at least one processor of a computing device comprising a two stage risk model, the relabeling based on a known status of a user associated with the at least one transaction;

weighting the plurality of transactions based on a type of each transaction in the plurality of transactions;

performing a first logistic regression on the weighted plurality of transactions to create the first stage of the two stage risk model;

performing a second equal-weighted logistic regression on transactions of known status generated by the first stage of the two stage risk model to generate the second stage of the two stage risk model; and

rejecting a specified percentage of resulting transactions.

10. The method of claim 8, wherein bad transactions are fraudulent transactions.

11. The method of claim 8, wherein the computerized transaction-based system is an order processing system.

12. The method of claim 9, further comprising:

evaluating the current risk model for total net profit generated for a randomly selected subset of a transaction population; and

evaluating the two stage risk model for total net profit generated for the randomly selected subset of the transaction population.

13. The method of claim 8, further comprising:

creating a plurality of instances of the two stage risk model using different sets of parameters; and

determining an instance of the plurality of instances that maximizes a measurable metric of the computerized transaction-based system.

14. A computer-readable storage medium comprising computer-readable instructions which when executed cause at least one processor of a computing device to:

evaluate a current model and a two stage risk model, the evaluation based on a measurable metric.

15. The computer-readable storage medium of claim 14, comprising further computer-readable instructions which when executed cause the at least one processor to:

randomly select a subset of a total transaction population generated by the current model;

solve for model building parameters which optimize net profit value associated with a first set of input parameters;

solve for model building parameters which optimize a net profit value associated with at least a second set of input parameters;

calculate the total net profit value generated by the current model for the subset;

calculate the total net profit value associated with the first set of input parameters;

calculate the total net profit value associated with the at least second set of input parameters;

determine a set of input parameters that maximizes net profit value; and

using the determined set of input parameters, solve for model building parameters.

16. The computer-readable storage medium of claim 14, comprising further computer-readable instructions which when executed cause the at least one processor to:

use the model building parameters to create a future model.

17. The computer-readable storage medium of claim 14, comprising further computer-readable instructions which when executed cause the at least one processor to:

receive a first parameter comprising a discount rate to account for multiple transactions placed by a good user whose transaction is rejected;

receive a second parameter comprising a discount rate to account for multiple transactions placed by a bad user whose transaction is rejected.

18. The computer-readable storage medium of claim 14, comprising further computer-readable instructions which when executed cause the at least one processor to:

instantiate a plurality of instances of the two stage risk model, wherein each instance of the plurality of instances is provided with a different set of first and second parameters;

determine the instance maximizing revenue, the revenue for each instance measured by comparing total net profit produced by each instance; and

use the set of first and second parameters associated with the instance that maximized revenue for a future model.

19. The computer-readable storage medium of claim 18, comprising further computer-readable instructions which when executed cause the at least one processor to:

adjust for retry transactions made by a first type of user with a first parameter and adjusting for retry transactions made by a second type of user with a second parameter; and

determine an instance of the plurality of instances of the two stage of risk model, the instance maximizing a specified metric comprising revenue measured by total net profit generated by each instance.

20. The computer-readable storage medium of claim 19, comprising further computer-readable instructions which when executed cause the at least one processor to:

evaluate the current model and the two stage risk model, the two stage risk model comprising a model for a transaction-based ordering system, the evaluation based on maximizing revenue.