US-12626256-B2 - Systems and methods for parallel hybrid model analysis

Abstract

A parallel hybrid model analysis system is provided. The system includes a memory device, and at least one processor coupled thereto. The processor programmed to: store a plurality of models including a first model of a first model type and a second model of a second model type wherein the first model type and the second model type are different; receive at least one data record to be analyzed; execute the first model with the at least one data record as input to receive one or more weight values as output; execute the second model with the at least one data record as input to receive a partial score as output; combine the one or more weight values with the partial score to generate a final score; and analyze the final score to determine if one or more alerts should be electronically transmitted to a user device.

Inventors

• Sayed Amin Afzal

Assignees

• MASTERCARD INTERNATIONAL INCORPORATED

Dates

Publication Date

20260512

Application Date

20231024

Claims (20)

1. 1 . A parallel hybrid model analysis (PHMA) system comprising: a PHMA server; a memory device; and at least one processor coupled to the memory device and in communication with the PHMA server, the at least one processor programmed to: store, at the PHMA server, a plurality of models including a first set of models of a first model type and a second model of a second model type, wherein the first model type is a clustering model type and the second model type is a neural network model type; receive at least one data record to be analyzed by the plurality of models; determine, based on one or more parameters of the at least one data record, which model of the first set of models to execute on the at least one data record; and cause the PHMA server to: execute the determined model of the first set of models with the at least one data record as input to (i) assign the at least one data record to a cluster associated with the determined model, including a cluster ID corresponding to the assigned cluster, and (ii) output one or more weight values associated with the cluster ID assigned to the at least one data record by the determined model, the determined model trained on data records associated with a first parameter option associated with the one or more parameters of the at least one data record; execute the second model with the at least one data record as input to output a partial score from the second model, the second model trained on data records associated with a second parameter option associated with the one or more parameters of the at least one data record; modify the partial score by (i) combining the one or more weight values with the partial score and (ii) adjusting a sensitivity of the partial score; generate a final score based on the modified partial score; analyze the final score in real-time, including comparing the final score to a score threshold to determine whether to electronically transmit one or more alerts associated with the score threshold to a user device; and based on a determination that the final score satisfies the score threshold, electronically transmit the one or more alerts to the user device.
2. 2 . The parallel hybrid model analysis system of claim 1 , wherein each model of the first set of models is associated with a respective option of a multi-option parameter of the one or more parameters of the at least one data record, and wherein the at least one processor is further programmed to cause the PHMA server to: determine, based on the one or more parameters of the at least one data record, which option of the multi-option parameter to associate with the at least one data record.
3. 3 . The parallel hybrid model analysis system of claim 2 , wherein the at least one processor is further programmed to cause the PHMA server to: preprocess the at least one data record including determining the one or more parameters of the at least one data record and formatting the at least one data record for input into the plurality of models.
4. 4 . The parallel hybrid model analysis system of claim 1 , wherein the second model is a second set of models and wherein the at least one processor is further programmed to: determine which model of the second set of models to execute based on the one or more parameters of the at least one data record.
5. 5 . The parallel hybrid model analysis system of claim 1 , wherein the clustering model type is a K-means clustering model and the cluster ID is associated with a cluster of the K-means clustering model.
6. 6 . The parallel hybrid model analysis system of claim 5 , wherein the one or more weight values are based on which cluster of the K-means clustering model the at least one data record is assigned to.
7. 7 . The parallel hybrid model analysis system of claim 1 , wherein the second model type is trained using a multilayer perception algorithm via supervised machine learning.
8. 8 . The parallel hybrid model analysis system of claim 1 , wherein to combine the one or more weight values with the partial score to generate a final score the at least one processor is further programmed to execute a sigmoid function.
9. 9 . The parallel hybrid model analysis system of claim 1 , wherein the at least one processor is further programmed to determine a reason code associated with the final score based at least in part on the one or more weight values, and the reason code is electronically transmitted as part of the one or more alerts to the user device.
10. 10 . The parallel hybrid model analysis system of claim 1 , wherein the at least one data record is a payment transaction, and at least one of the first set of models and the second model is trained to detect a potential money laundering scheme that includes the at least one data record.
11. 11 . A computer-implemented method for parallel hybrid model analysis (PHMA), the computer-implemented method implemented on a computing device comprising a memory device coupled to at least one processor, the at least one processor being in communication with a PHMA server, the computer-implemented method comprising: storing, at the PHMA server, a plurality of models including a first set of models of a first model type and a second model of a second model type, wherein the first model type is a clustering model type and the second model type is a neural network model type; receiving at least one data record to be analyzed by the plurality of models; determining, based on one or more parameters of the at least one data record, which model of the first set of models to execute on the at least one data record; executing, via the PHMA server, the determined model of the first set of models with the at least one data record as input to (i) assign the at least one data record to a cluster associated with the determined model, including a cluster ID corresponding to the assigned cluster, and (ii) output one or more weight values associated with the cluster ID assigned to the at least one data record by the determined model, the determined model trained on data records associated with a first parameter option associated with the one or more parameters of the at least one data record; executing, via the PHMA server, the second model with the at least one data record as input to output a partial score from the second model, the second model trained on data records associated with a second parameter option associated with the one or more parameters of the at least one data record; modify the partial score by (i) combining, via the PHMA server, the one or more weight values with the partial score and (ii) adjusting a sensitivity of the partial score; generating a final score based on the modified partial score; analyzing, via the PHMA server, the final score in real-time, including comparing, via the PHMA server, the final score to a score threshold to determine whether to electronically transmit one or more alerts associated with the score threshold to a user device; and based on a determination that the final score satisfies the score threshold, electronically transmitting, via the PHMA server, the one or more alerts to the user device.
12. 12 . The computer-implemented method of claim 11 , wherein each model of the first set of models is associated with a respective option of a multi-option parameter of the one or more parameters of the at least one data record, the computer-implemented method further comprising: determining, based on the one or more parameters of the at least one data record, which option of the multi-option parameter to associate with the at least one data record.
13. 13 . The computer-implemented method of claim 12 , further comprising preprocessing, via the PHMA server, the at least one data record including determining the one or more parameters of the at least one data record and formatting the at least one data record for input into the plurality of models.
14. 14 . The computer-implemented method of claim 11 , wherein the second model is a second set of models, the computer-implemented method further comprising: determining which model of the second set of models to execute based on the one or more parameters of the at least one data record.
15. 15 . The computer-implemented method of claim 11 , wherein the clustering model type is a K-means clustering model and the cluster ID is associated with a cluster of the K-means clustering model.
16. 16 . The computer-implemented method of claim 15 , wherein the one or more weight values are based on which cluster of the K-means clustering model the at least one data record is assigned to.
17. 17 . The computer-implemented method of claim 11 , wherein the second model type is trained using a multilayer perception algorithm via supervised machine learning.
18. 18 . The computer-implemented method of claim 11 further comprising determining a reason code associated with the final score based at least in part on the one or more weight values and electronically transmitting the reason code as part of the one or more alerts to the user device.
19. 19 . The computer-implemented method of claim 11 , wherein the at least one data record is a payment transaction, and at least one model of the first set of models and the second model is trained to detect a potential money laundering scheme that includes the at least one data record.
20. 20 . At least one non-transitory computer-readable storage media having computer-executable instructions embodied thereon for parallel hybrid model analysis (PHMA), wherein when executed by at least one processor of a PHMA system, the at least one processor being in communication with a PHMA server, the computer-executable instructions cause the at least one processor to: store, at the PHMA server, a plurality of models including a first set of models of a first model type and a second model of a second model type, wherein the first model type is a clustering model type and the second model type is a neural network model type; receive at least one data record to be analyzed by the plurality of models; determine, based on one or more parameters of the at least one data record, which model of the first set of models to execute on the at least one data record; and cause the PHMA server to: execute the determined model of the first set of models with the at least one data record as input to (i) assign the at least one data record to a cluster associated with the determined model, including a cluster ID corresponding to the assigned cluster, and (ii) output one or more weight values associated with the cluster ID assigned to the at least one data record by the determined model, the determined model trained on data records associated with a first parameter option associated with the one or more parameters of the at least one data record; execute the second model with the at least one data record as input to output a partial score from the second model, the second model trained on data records associated with a second parameter option associated with the one or more parameters of the at least one data record; modify the partial score by (i) combining the one or more weight values with the partial score and (ii) adjusting a sensitivity of the partial score; generate a final score based on the modified partial score; analyze the final score in real-time, including comparing the final score to a score threshold to determine whether to electronically transmit one or more alerts associated with the score threshold to a user device; and based on a determination that the final score satisfies the score threshold, electronically transmit the one or more alerts to the user device.

Description

BACKGROUND The present application relates generally to parallel hybrid model analysis, and more particularly, to computer-based systems and methods for using multiple machine learning models in parallel. In at least some data analysis systems, a model may be trained using machine learning techniques to detect patterns within the data. However, as the number of data records analyzed and the number of variables to be tracked increases, the difficulty of training such a model also increases, in some cases exponentially. Also, using the known systems for data analysis, the more complicated the analysis, the more computational resources may be needed to perform the analysis. Generating and training a single model that can address all of the potential details in analyzing large numbers of data records may be overly large, complicated, and difficult to use and may require significant computing resources to execute. Furthermore, adding the capability to handle new details may be difficult and the single model may become unwieldy to use. These large models can also lack the needed accuracy and granularity in producing results. Conventional techniques may have additional encumbrances, inefficiencies, ineffectiveness, and drawbacks as well. Accordingly, a more resource efficient system and/or method for data analysis systems would be desirable. BRIEF DESCRIPTION In one aspect, a parallel hybrid model analysis system is provided. The parallel hybrid model analysis system includes a memory device and at least one processor coupled to the memory device. The at least one processor is programmed to store a plurality of models including a first model of a first model type and a second model of a second model type. The first model type and the second model type are different. The at least one processor is also programmed to receive at least one data record to be analyzed. The at least one processor is further programmed to execute the first model with the at least one data record as input to receive one or more weight values as output. In addition, the at least one processor is programmed to execute the second model with the at least one data record as input to receive a partial score as output. Moreover, the at least one processor is programmed to combine the one or more weight values with the partial score to generate a final score. Furthermore, the at least one processor is programmed to analyze the final score to determine if one or more alerts should be electronically transmitted to a user device. In another aspect, a computer-implemented method for parallel hybrid model analysis is provided. The method is implemented on a computing device comprising a memory device coupled to at least one processor. The method includes storing a plurality of models including a first model of a first model type and a second model of a second model type. The first model type and the second model type are different. The method also includes receiving at least one data record to be analyzed. The method further includes executing the first model with the at least one data record as input to receive one or more weight values as output. In addition, the method includes executing the second model with the at least one data record as input to receive a partial score as output. Moreover, the method includes combining the one or more weight values with the partial score to generate a final score. Furthermore, the method includes analyzing the final score to determine if one or more alerts should be electronically transmitted to a user device. In a further aspect, at least one non-transitory computer-readable storage media having computer-executable instructions embodied thereon for authenticating an online user is provided. When executed by at least one processor, the computer-executable instructions cause the at least one processor to store a plurality of models including a first model of a first model type and a second model of a second model type. The first model type and the second model type are different. The computer-executable instructions also cause the at least one processor to receive at least one data record to be analyzed. The computer-executable instructions further cause the at least one processor to execute the first model with the at least one data record as input to receive one or more weight values as output. In addition, the computer-executable instructions cause the at least one processor to execute the second model with the at least one data record as input to receive a partial score as output. Moreover, the computer-executable instructions cause the at least one processor to combine the one or more weight values with the partial score to generate a final score. Furthermore, the computer-executable instructions further cause the at least one processor to analyze the final score to determine if one or more alerts should be electronically transmitted to a user device. BRIEF DESCRIPTION OF THE DRAWINGS FIGS. 1-6 show example e