System and method of detecting and assessing multiple types of risks related to mortgage lending

• US 8,489,499 B2
• Filed: 01/13/2010
• Issued: 07/16/2013
• Est. Priority Date: 01/13/2010
• Status: Active Grant

• Current Assignee: Corelogic Solutions LLC (CoreLogic Incorporated)
• Original Assignee: Corelogic Solutions LLC (CoreLogic Incorporated)
• Inventors: Yan, Rui, Chi, Hoi-Ming, Koo, Seongjoon, Baker, James
• Primary Examiner(s): Shrestha, Bijendra K
• Application Number:
  US12/687,098
• Publication Number:
  US 20110173116A1
• Time in Patent Office: 1,280 Days
• Field of Search: 705/38
• US Class Current: 705/38
• CPC Class Codes: G06N 20/10 using kernel methods, e.g. support vector machines [SVM]
• CPC Class Codes: G06N 20/20 Ensemble learning
• CPC Class Codes: G06N 3/084 Backpropagation, e.g. using gradient descent
• CPC Class Codes: G06N 3/088 Non-supervised learning, e.g. competitive learning
• CPC Class Codes: G06N 3/126 Evolutionary algorithms, e.g. genetic algorithms or genetic programming
• CPC Class Codes: G06N 5/01 Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
• CPC Class Codes: G06N 5/048 Fuzzy inferencing
• CPC Class Codes: G06Q 10/067 Enterprise or organisation modelling
• CPC Class Codes: G06Q 30/0185 Product, service or business identity fraud
• CPC Class Codes: G06Q 40/02 Banking, e.g. interest calculation or account maintenance credit or loans G06Q40/03
• CPC Class Codes: G06Q 40/03 Credit; Loans; Processing thereof

Abstract

Embodiments include systems and methods of detecting and assessing multiple types of risks related to mortgage lending. One embodiment includes a system and method of detecting and assessing risks including fraud risks, early payment default risks, and risks related to fraudulently stated income on loan applications. One embodiment includes a computerized method that includes creating a combined risk detection model based on a plurality of risk detection models and using the combined risk detection model to evaluate loan application data and generate a combined risk score that takes into account interaction of different types of risks individually and collectively detected by the plurality of risk detection models.

Claims

1. A system for detecting and assessing lending risks, the system comprising:

a computer system comprising one or more computing devices, the computer system programmed, via executable code modules, to implement;

a combined risk detection model for detecting and assessing data indicative of a plurality of risks in loan application data, the combined risk detection model adapted to receive as input a plurality of input features extracted from two or more of a plurality of risk detection models, the plurality of risk detection models comprising;

a fraud model that detects the presence of data indicative of fraud in the loan application data, the fraud model configured to generate a fraud model score that is indicative of the presence of fraud in the loan application data;

a multi-component risk model that assesses risks associated with a loan referenced by the loan application data, the multi-component risk model based at least in part on external data not within the loan application data, the multi-component risk model configured to generate a multi-component risk score indicative of risk associated with the loan; and

a default risk model that detects the presence of data indicative of a risk of early payment default in the loan application data, the default risk model configured to generate a default risk score indicative of the risk of early payment default,

wherein the computer system is further programmed to;

determine combinability of the plurality of risk detection models based at least partly on evaluating predictive performance of scores from one or more combinations of the risk detection models against historical transactions data; and

extracting input features from the risk detection models that are determined to be combinable, the input features created from a mathematical combination of the scores and additional data, wherein the scores from the models determined to be combinable include one or more of the fraud model score, the multi-component risk score, or the default risk score;

generate a composite risk score based at least in part on the scores from the models determined to be combinable and the extracted input features; and

a score reporting module that reports the composite risk score generated by the combined risk detection model.

2. The system of claim 1, wherein the input features are extracted from one or more of:

scores from the plurality of risk detection models or additional data related to the loan application not output by the plurality of risk detection models.

3. The system of claim 2, wherein the additional data related to the loan application not output by the plurality of risk detection models comprises a loan amount.

4. The system of claim 1, wherein the score reporting module further reports a plurality of risk indicators generated by the plurality of risk detection models.

5. The system of claim 4, wherein each of the risk indicators references a loan risk and is classified in accordance with a weight contribution of the referenced loan risk to the composite risk score.

6. The system of claim 5, wherein each of the risk indicators is classified as a high risk, a moderate risk, or a low risk based on the weight contribution of the referenced loan risk.

7. The system of claim 1, wherein a modeling method used to construct the combined risk detection model comprises one of:

linear regression, logical regression, neural networks, support vector machines, or decision trees generated using a machine learning algorithm that uses a tree-like graph to predict an outcome.

8. The system of claim 1, wherein the combined risk detection model comprises one or more of the following modeling structures:

a cascade structure in which scores of the risk detection models are ranked in a specified order, and at least a first risk detection model score is used to generate a first stage score, and at least the first stage score and a second risk detection score are used to generate a second stage score; and

a divide-and-conquer structure in which a combination gate incorporates scores from the plurality of risk detection models with interactive input fields to generate the composite risk score.

9. The system of claim 1, wherein the evaluating of the predictive performance of the risk detection models includes a correlation analysis or a swap analysis of the scores generated by the risk detection models, wherein the correlation analysis determines similarity among the scores from the risk detection models and the swap analysis determines overlap of a review population based on outsorting logic of the scores of the risk detection models.

10. The system of claim 1, wherein the input features correspond to results of mathematical operations on the scores from the plurality of risk detection models, the mathematical operations further comprising:

obtaining a maximum of the scores, obtaining a minimum of the scores, obtaining an average of the scores, and obtaining a dynamic range of the scores, and obtaining a ratio of the dynamic range over the average.

11. The system of claim 1, wherein the multi-component risk model outputs a plurality of component scores comprising one or more of:

a collateral component score, a broker component score, a borrower component score, and a market component score.

12. The system of claim 11, wherein the one or more component scores are calculated based on external data not within the loan application data, the external data comprising data contributed by lenders, credit data, and public records data.

27. The system of claim 1, wherein the input features are extracted from the two or more risk detection models by mathematically combining scores from the plurality of risk detection models for input into the combined risk detection model, the input features being selected as based at least in part on a comparison of the predictive performance of the risk detection models and a selection of a modeling method used to construct the combined risk detection model.

28. The system of claim 1, wherein the plurality of risk detection models further comprises an income fraud detection model that detects the presence of data indicative of fraud in income data within the mortgage transaction data, the income fraud detection model configured to generate an income fraud score indicative of fraud in the income data.

13. A computerized method of detecting and assessing mortgage lending risks, the method comprising:

receiving, on a physical computer processor, mortgage transaction data that includes loan application data and historical transactions data;

determining, on a physical computer processor, combinability of a plurality of risk detection models, the determining comprising determining predictive performance of scores from one or more combinations of the risk detection models as compared to the historical transactions data, wherein the plurality of risk detection models further comprise two or more of;

an income fraud detection model that detects the presence of data indicative of fraud in income data within the mortgage transaction data, the income fraud detection model generating an income fraud score indicative of fraud in the income data;

a default risk model that detects a risk of payment default in the mortgage transaction data, the default risk model generating a default risk score indicative of the risk of payment default; and

a multi-component risk model that detects risks in loan transactions, the multi-component risk model further comprising scores from one or more of the following components;

a collateral component, a broker component, a borrower component, and a market component, the

multi-component risk model generating a multi-component risk score indicative of risk in the mortgage transaction data;

extracting, on a physical computer processor, input features from the risk detection models that are determined to be combinable, the input features created from a mathematical combination of the scores and additional data;

applying, on a physical computer processor, a combined risk detection model to the mortgage transaction data to generate a composite risk score, wherein the combined risk detection model is configured to receive the extracted input features from the plurality of risk detection models determined to be combinable and mortgage transaction data not output by the risk detection models, the extracted input features further being selected as based at least in part on a modeling method used to construct the combined risk detection model;

generating, on a physical computer processor, a report including the composite risk score generated by the combined risk detection model.

14. The method of claim 13, determining the combinability of the plurality of risk detection models is based at least in part on the correlation of the results of applying the plurality of risk detection models to historical transaction data.

15. The method of claim 14, wherein the correlation is based at least in part on a measure of the similarity of the results among the plurality of risk detection models.

16. The method of claim 13, wherein the modeling method is one of:

linear regression, logical regression, neural networks, support vector machines, and decision trees generated using a machine learning algorithm that uses a tree-like graph to predict an outcome.

17. The method of claim 13, wherein the input features are selected by:

applying each of the plurality of risk detection models to data to generate a score for each risk detection model, the data comprising historical mortgage transaction data;

identifying an interaction among scores from the plurality of risk detection models; and

using the interaction as a basis for the selection of the input features.

18. The method of claim 13, wherein the input features are selected by:

applying each of the plurality of risk detection models to data to generate a score for each risk detection model, the data comprising historical mortgage transaction data;

performing a swap analysis on the scores from applying the plurality of risk detection models to the data; and

using the result of the swap analysis as a basis for the selection of the input features.

19. The method of claim 13, wherein the generating further provides a plurality of risk indicators generated by the combined risk detection model.

20. The method of claim 19, wherein each of the risk indicators references a loan risk and is classified in accordance with a weight contribution of the referenced loan risk to the composite risk score.

21. The method of claim 20, wherein each of the risk indicators is classified as a high risk, a moderate risk, or a low risk based on the weight contribution of the referenced loan risk.

22. The method of claim 13, wherein the default risk model detects a risk of payment default occurring in one of the following periods:

within the first 90 days of a loan repayment period, within the first six months of the loan repayment period, and within the first year of the loan repayment period.

23. A method for creating a model for detecting loan risks, the method comprising:

receiving loan transaction data linked with historical loan performance data related to risks including fraud and default risks;

executing a plurality of risk detection models on the loan transaction data to obtain a respective model score from each of a plurality of risk detection models, wherein the plurality of risk detection models further comprise two or more of;

a fraud model that detects the presence of data indicative of fraud, the fraud model derived from loan transaction data, the fraud model generating a fraud score indicative of the presence of fraud in the loan transaction data;

a multi-component risk model that detects risks in loan transactions, the multi-component risk model based at least in part on data external to the loan application data including credit data and public records data, the multi-component risk model generating a multi-component risk score indicative of the risk in the loan transaction data; and

a default risk model that detects the presence of data indicative of a risk of early payment default, the default risk model generating a default risk score indicative of the risk of early payment default;

determining the combinability of the plurality of risk detection models based at least partly on evaluating predictive performance of scores from one or more combinations of the risk detection models against the historical loan performance data;

extracting input features from the risk detection models that are determined to be combinable, the input features created from a mathematical combination of the scores and additional data, wherein the scores from the risk detection models determined to be combinable include one or more of the fraud score, the multi-component risk score, or the default risk score; and

constructing a combined risk detection model based on the extracted input features,

the receiving, executing, determining, extracting, and constructing performed by a computer system that comprises one or more computing devices.

24. The method of claim 23, wherein the constructing further comprises:

selecting one or more model structures for the combined risk detection model;

selecting a modeling method for the combined risk detection model;

training the combined risk detection model on the loan transaction data; and

evaluating the performance of the combined risk detection model, the selecting one or more model structures, selecting a modeling method, training, and evaluating performed by a computer system that comprises one or more computing devices.

25. The method of claim 24, wherein the modeling method is one of:

linear regression, logical regression, neural networks, support vector machines, and decision trees generated using a machine learning algorithm that uses a tree-like graph to predict an outcome.

26. The method of claim 24, wherein the one or more model structures comprises:

a cascade structure in which scores of the risk detection models are ranked in a specified order, and at least a first risk detection model score is used to generate a first stage score, and at least the first stage score and a second risk detection score are used to generate a second stage score; and

a divide-and-conquer structure in which a combination gate incorporates scores from the plurality of risk detection models with interactive input fields to generate the composite risk score.