Real-time anti-fraud detection system for digital financial transaction

By designing a real-time anti-fraud detection system in the digital financial transaction system, and using machine learning algorithms and SMOTE technology to build an integrated classification model, the problem of identifying complex digital financial fraud is solved, and effective identification and response to new types of fraud is achieved.

CN119090518AInactive Publication Date: 2024-12-06NANTONG YAOER LINGJIU INFORMATION TECH CO LTD
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411229134.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-03
Publication Date
2024-12-06
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

It is difficult for the prior art to effectively identify and deal with complex digital financial fraud, especially because fraudulent behavior accounts for a small proportion of transaction behavior, similar characteristics to normal samples, and the fraud methods are constantly changing, resulting in classifiers being unable to adapt to new feature rules.

Method used

Design a real-time anti-fraud detection system for digital financial transactions, including fraud detection module, fraud processing module, real-time analysis module, risk warning module and feedback learning module, and generate fraud samples using machine learning algorithms and SMOTE technology, combining multiple classifiers and Feature analysis, build integrated classification models, and continuously optimize the models through real-time analysis and feedback learning.

Benefits of technology

Improve model performance, reduce the impact of unbalanced data, effectively identify and deal with new fraud, and improve detection accuracy and system adaptability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119090518A_ABST
    Figure CN119090518A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of financial monitoring, and particularly discloses a digital financial transaction real-time anti-fraud detection system, which comprises a fraud detection module, a fraud processing module, a real-time analysis module and a risk early warning module, and is characterized in that the fraud detection module analyzes suspicious transactions by using a machine learning algorithm to identify and prevent potential fraud activities; the fraud processing module automatically intercepts and processes transactions with fraud behaviors and submits suspicious events to manual auditing, and the real-time analysis module is used for performing real-time analysis on transaction data and identifying and marking suspicious transactions in time; according to the method, a plurality of classifiers are constructed by using a machine learning algorithm, a training sample and classifier parameter combination is obtained by using an SMOTE resampling mode and AUROC estimation, the plurality of constructed classifiers are trained, then the classifiers are used for constructing a detection model of integrated classification results, digital financial transaction data analyzed in real time are detected, the model performance is improved, and the detection accuracy is improved. And the influence caused by unbalanced data is reduced.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of financial monitoring, and in particular relates to a real-time anti-fraud detection system for digital financial transactions. Background Art

[0002] Digital finance is a new financial model that integrates digitization, informatization and traditional financial industries. The rapid development of new generation information technology, on the one hand, empowers the financial industry with science and technology, giving birth to a new form of digital finance, which can effectively improve the efficiency of financial services and reduce operating costs; on the other hand, digital technology reacts to the financial field and also triggers a series of financial risks. New financial fraud incidents that are frequent, highly concerned and involve a wide range of people have seriously disrupted the normal financial order and hindered the high-quality development of the social economy.

[0003] With the development of digital finance, financial fraud has evolved from simple hacking and theft to high-precision digital financial fraud with the help of information technology. The fraud methods are complex and diverse and difficult to identify. The main forms include loan default fraud and credit card fraud in the traditional financial field and online loan fraud, Internet credit fraud, virtual currency money laundering, etc. in the digital financial field. In the process of real-time detection of whether there is fraud, there are two main difficulties: first, fraud only accounts for a small part of all transactions, the data category distribution is highly unbalanced, and the feature information of some fraud samples and normal samples is highly similar, making it difficult for ordinary classification methods to effectively identify such fraud. Secondly, fraud is time-sensitive. Since fraud platforms or fraud gangs will constantly change their methods in response to anti-fraud measures, the classifier cannot adapt to the new feature rules, which in turn makes it impossible to effectively identify fraud.

[0004] Therefore, it is necessary to propose a real-time anti-fraud detection system for digital financial transactions to at least partially solve the problems existing in the prior art.

[0005] The above information disclosed in this background technology is only used to increase the understanding of the background technology of the present invention and therefore, it may include information that does not constitute the prior art known to ordinary technicians in this field. Summary of the invention

[0006] The object of the present invention is to provide a real-time anti-fraud detection system for digital financial transactions to solve the problems raised in the above-mentioned background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A real-time anti-fraud detection system for digital financial transactions comprises: a fraud detection module, a fraud processing module, a real-time analysis module, a risk warning module, a transaction data collection module, and a feedback learning module. The fraud detection module uses a machine learning algorithm to analyze suspicious transactions to identify and prevent potential fraudulent activities. The fraud processing module automatically intercepts and processes transactions involving fraudulent behavior and submits suspicious events for manual review. The real-time analysis module is used to perform real-time analysis on transaction data, timely identify and mark suspicious transactions, and the risk warning module automatically sends warning information to participants based on the analysis results. The transaction data collection module obtains digital financial transaction behavior data in real time, integrates the data and stores it in a system database. The feedback learning module continuously adjusts and optimizes the model based on actual detection results and user feedback to improve detection accuracy and system adaptability.

[0009] Preferably, the fraud detection module uses a machine learning algorithm to construct multiple classifiers, uses SMOTE technology to generate samples with fraudulent behavior, combines the acquired transaction behavior data to construct a training data set to train the classifier, and then uses the classifier to construct a detection model that integrates the classification results for detecting new transaction data.

[0010] The original data set to be trained is segmented based on the K-fold crossover principle, and then a variety of different combinations of classifier parameter sets are prepared. The currently segmented training set is resampled by SMOTE, and all new samples generated by K rounds are integrated. The samples that have completed the resampling process are trained using different parameter combinations, and the new sample batches and the current parameter combinations are recorded. Finally, the training model under the current new samples and different training parameter combinations is obtained. The SMOTE resampling steps are as follows:

[0011] (1) Calculate several neighboring sample sets of the current sample through the KNN algorithm;

[0012] (2) Randomly select a sample from the set and calculate the difference between it and all the feature values ​​of the current sample;

[0013] (3) Randomly generate a number rand in the range of 0 to 1;

[0014] (4) Synthesize a new minority class sample based on the current sample.

[0015] Repeat the process of generating and training new samples to obtain training models with multiple synthetic samples and parameter combinations, sort the AUROC values ​​of all model results, and select the top several combinations with the highest AUROC scores (determine the selected combinations based on actual needs). Train multiple classifiers with 1 / K-1 times the number of samples selected by sampling and the corresponding parameter combinations. Receive the trained classifiers and use them to classify the test sets respectively, obtain the classification tendency value of each model for each sample to be tested, and perform weighted average on multiple tendency values ​​of each sample to be tested, and finally obtain the classification tendency value of all samples to be tested, that is, the classification result.

[0016] Preferably, the fraud processing module uses a rule engine to perform analysis based on the detection results, automatically intercepts transactions involving fraudulent behavior, and classifies suspicious events by type to help manual auditors better understand and handle problems, and automatically generates a detailed fraud report after intercepting or handling the problem.

[0017] Preferably, the real-time analysis module constructs a feature set based on the acquired transaction behavior data using multi-fractal analysis and firefly algorithm, uses the Pribot model to eliminate factors that are not significantly correlated with digital financial fraud in the feature set, and then combines it with the features selected by artificial prior knowledge to obtain a key feature set to provide a high-quality data foundation for subsequent classification.

[0018] Based on the acquired transaction behavior data, multifractal analysis is used as the evaluation criterion to determine the number of features in the feature set. The firefly algorithm is used as a search method to quickly search for the initial feature set that is closest to the multifractal analysis. The selected feature set is trained using a classifier to evaluate the classification performance of the feature set. Then, based on the knowledge of digital financial fraud, the Pribot model is used to eliminate factors that are not significantly related to the feature set and digital financial fraud.

[0019] Preferably, the risk warning module generates warning information containing detailed information according to the event type, and then sends it to the participants through multiple notification channels, tracks and records the response to the alarm, and regularly evaluates the effectiveness of the warning for subsequent analysis and improvement.

[0020] Preferably, the transaction data collection module uses the AES encryption algorithm in GCM mode to encrypt the digital financial transaction behavior data, and implements interaction with the database through the LabSQL toolkit using Microsoft ADO technology and SQL language.

[0021] Preferably, the feedback learning module uses user feedback as new training data, learns the characteristics of the new data through an online learning algorithm to update the detection model, and uses One-Class SVM anomaly detection to detect new and unseen fraud patterns.

[0022] A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the processor implements a real-time anti-fraud detection system for digital financial transactions as described above.

[0023] A terminal device includes a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, wherein the instructions are suitable for being loaded by the processor and executed by a real-time anti-fraud detection system for digital financial transactions as described above.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] The present invention uses a machine learning algorithm to construct multiple classifiers, utilizes the SMOTE resampling method and AUROC valuation to obtain a combination of training samples and classifier parameters, trains the constructed multiple classifiers, and then uses the classifiers to construct a detection model that integrates the classification results to detect real-time analyzed digital financial transaction data, improve model performance, and reduce the impact of unbalanced data. By performing importance decay sampling on historical sample data, new knowledge rules or concepts are valued and applied in the detection model, thereby being able to deal with new fraudulent behaviors.

[0026] The above summary is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments and features described above, further aspects, embodiments and features of the present invention will be readily apparent by reference to the accompanying drawings and the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 A framework diagram of a real-time anti-fraud detection system for digital financial transactions according to the present invention. DETAILED DESCRIPTION

[0028] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] It should be further understood that the term “and / or” used in the specification of this application and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0030] Embodiment 1:

[0031] See also Figure 1 As shown, a real-time anti-fraud detection system for digital financial transactions includes: a fraud detection module, a fraud processing module, a real-time analysis module, a risk warning module, a transaction data collection module, and a feedback learning module. The fraud detection module uses a machine learning algorithm to analyze suspicious transactions to identify and prevent potential fraudulent activities. The fraud processing module automatically intercepts and processes transactions involving fraudulent behavior, and submits suspicious events for manual review. The real-time analysis module is used to perform real-time analysis on transaction data, promptly identify and mark suspicious transactions, and the risk warning module automatically sends warning information to participants based on the analysis results. The transaction data collection module obtains digital financial transaction behavior data in real time, integrates the data and stores it in the system database. The feedback learning module continuously adjusts and optimizes the model based on actual detection results and user feedback to improve detection accuracy and system adaptability.

[0032] The fraud detection module uses machine learning algorithms to analyze suspicious transactions to identify and prevent potential fraudulent activities. It uses machine learning algorithms to build multiple classifiers, uses SMOTE technology to generate samples of fraudulent behavior, and combines the acquired transaction behavior data to build a training data set to train the classifier. The classifier is then used to build a detection model that integrates the classification results for detecting new transaction data.

[0033] When using sample data for training, we take into account that as time goes by, the knowledge rules or concepts contained in the data will change accordingly, new knowledge rules or concepts will be added, and they will be more important than the original knowledge rules or concepts. Therefore, sample selection is completed in accordance with the law that the importance of knowledge of historical samples decreases over time. That is, the older the historical samples are, the smaller the number of training samples extracted from them.

[0034] Example: Split the original data set into 20 data blocks in units of 3 days, and each data block corresponds to the sample set that arrives in each time period. The first 3-day data block will be used as the initial training sample set, and the second 3-day data block will be used as the first test set. When the third 3-day data block arrives, the second data block will be merged with the first data block as a candidate training sample set, and the same method will be used for subsequent data blocks.

[0035] The training sample set selected by the model for each time unit is T 1-1 , the calculation formula is as follows:

[0036]

[0037] In the formula, w represents the current time unit, t represents the historical time unit, T is the dataset of training samples, S t Represents the sample set required to be classified in t time units, Indicates that from St Sample set extraction In actual scenarios, the sample set still has a high degree of category imbalance, so the above SMOTE resampling process is added before each training to ensure that the category distribution of the training set in each stage of model learning is balanced, thereby reducing the impact of imbalanced data.

[0038] The fraud processing module automatically intercepts and processes transactions involving fraudulent behavior, submits suspicious events for manual review, uses the rule engine to analyze based on the detection results, automatically intercepts transactions involving fraudulent behavior, and classifies suspicious events by type to help manual auditors better understand and handle problems, and automatically generates detailed fraud reports after intercepting or handling problems.

[0039] The real-time analysis module is used to perform real-time analysis on transaction data, promptly identify and mark suspicious transactions, and construct a feature set based on the acquired transaction behavior data using multi-fractal analysis and the firefly algorithm. The Pribot model is used to eliminate factors that are not significantly correlated with digital financial fraud in the feature set, and then it is combined with the features selected by artificial prior knowledge to obtain the key feature set, providing a high-quality data foundation for subsequent classification.

[0040] Use the optimal point set to initialize the population so that the firefly population is evenly distributed in the solution space. m is a unit cube in m-dimensional Euclidean space, the initial population size is n, let r∈G m , then the same as p n (j) = {((r 1 ×j),(r 2 ×j),…,(r m ×j)),j=1,2,…,n} is called the good point set, which means taking n good points uniformly designed from the θ-dimensional space, using the exponential sequence: r i ={e i ,1≤i≤m} to initialize the firefly population.

[0041] Assume that at the tth iteration, the current firefly X a =(x a1 ,x a2 ,…,x ak ), target Firefly X b =(x b1 ,x b2 ,…,x bk ), k is the dimension of the solution vector, and X a and X b The distance between them is used as the moving step, and the current firefly is randomly changed. Bit, that is, select the s bit of the current firefly, change it from 1 to 0 or from 0 to 1, and update the position.

[0042] X a and X b The distance h ab The calculation formula is as follows:

[0043]

[0044] In the formula, h abl Represents X a and X b The distance in the lth dimension.

[0045] In the process of spatial diffusion, with the parent generation as the axis, the offspring diffuses according to the law of normal distribution, and the maximum and minimum standard deviations are distributed as λ 1 and λ 0 , the number of iterations is ω, and the maximum number of iterations is ω max , α is the nonlinear harmonic index.

[0046]

[0047] Introducing the tangent function, the standard deviation is as follows:

[0048]

[0049] In the initial stage, a larger λ is selected, and the global search ability is stronger, so that offspring that are farther away from the parent are obtained. If the optimal offspring individual produced is better than the parent, the current firefly is updated; otherwise, it is not updated.

[0050] Introduce a mutation mechanism to expand the search range of individual fireflies in the neighborhood to enrich the population diversity. Introduce a mutation factor γ. If rand < γ, randomly change the neighbor position of the current firefly, changing the selected neighbor position from 0 to 1 or from 1 to 0. In this way, the fireflies can mutate with a certain probability, increase the randomness of individual searches, and make them jump out of the local optimum.

[0051] Embodiment 2:

[0052] The risk warning module automatically sends warning information to participants based on the analysis results, generates warning information containing detailed information according to the event type, and then sends it to participants through multiple notification channels, tracks and records the response to the alarm, and regularly evaluates the effectiveness of the warning for subsequent analysis and improvement.

[0053] When the risk warning module generates warning information, the membership degree and grayscale are integrated through the gray fuzzy comprehensive evaluation method to target the uncertainty of the dangerous degree of fraudulent behavior. The membership degree is used to measure the quality of qualitative indicators, and the grayscale is used to describe the insufficiency of indicator information. The gray system theory is used to obtain gray statistics, and a gray fuzzy matrix is ​​constructed. Finally, the gray fuzzy algorithm is used to obtain the dangerous degree of fraud risk, providing data support for the generation of warning information.

[0054] The transaction data collection module acquires digital financial transaction behavior data in real time, integrates the data and stores it in the system database, encrypts the digital financial transaction behavior data using the AES encryption algorithm in GCM mode, and interacts with the database through the LabSQL toolkit using Microsoft ADO technology and SQL language.

[0055] The feedback learning module continuously adjusts and optimizes the model according to the actual detection results and user feedback to improve the detection accuracy and the adaptability of the system. It uses user feedback as new training data and learns the characteristics of new data through online learning algorithms to update the detection model. It uses One-Class SVM anomaly detection to detect new and unseen fraud patterns.

[0056] From the above, it can be seen that the present invention uses a machine learning algorithm to construct multiple classifiers, uses the SMOTE resampling method and AUROC valuation to obtain a combination of training samples and classifier parameters, trains and constructs multiple classifiers, and then uses the classifiers to build a detection model that integrates the classification results to detect real-time analyzed digital financial transaction data, improve model performance, and reduce the impact of unbalanced data. By performing importance decay sampling on historical sample data, new knowledge rules or concepts are valued and applied in the detection model, thereby being able to deal with new fraudulent behaviors.

[0057] Embodiment 3:

[0058] An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the processor implements a real-time anti-fraud detection system for digital financial transactions as described above.

[0059] An embodiment of the present invention also provides a terminal device, including a processor and a computer-readable storage medium, the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, and the instructions are suitable for being loaded and executed by the processor as a real-time anti-fraud detection system for digital financial transactions as described above.

[0060] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0061] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to generate a program, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A way to specify functions in one or more boxes.

[0062] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0063] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0064] In the drawings of the embodiments disclosed in the present invention, only the structures involved in the embodiments disclosed in the present invention are involved, and other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of the present invention can be combined with each other.

[0065] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0066] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A real-time anti-fraud detection system for digital financial transactions, comprising: The fraud detection module, the fraud processing module, the real-time analysis module, and the risk warning module are characterized in that the fraud detection module uses a machine learning algorithm to analyze suspicious transactions to identify and prevent potential fraudulent activities; the fraud processing module automatically intercepts and processes transactions involving fraudulent behavior and submits suspicious events for manual review; the real-time analysis module is used to perform real-time analysis of transaction data, timely identify and mark suspicious transactions, and the risk warning module automatically sends warning information to participants based on the analysis results.

2. A digital financial transaction real-time anti-fraud detection system according to claim 1, characterized in that: The fraud detection module uses a machine learning algorithm to build multiple classifiers, uses SMOTE technology to generate samples with fraudulent behavior, combines the acquired transaction behavior data to build a training data set to train the classifier, and then uses the classifier to build a detection model that integrates the classification results for detecting new transaction data.

3. A digital financial transaction real-time anti-fraud detection system according to claim 2, characterized in that: The fraud processing module uses a rule engine to perform analysis based on the detection results, automatically intercepts transactions involving fraudulent behavior, and classifies suspicious events by type to help manual auditors better understand and handle problems, and automatically generates detailed fraud reports after intercepting or handling problems.

4. A digital financial transaction real-time anti-fraud detection system according to claim 3, characterized in that: The real-time analysis module constructs a feature set based on the acquired transaction behavior data using multi-fractal analysis and firefly algorithm, uses the Pribot model to eliminate factors that are not significantly correlated with digital financial fraud, and then combines it with the features selected by artificial prior knowledge to obtain a key feature set to provide a high-quality data foundation for subsequent classification.

5. A digital financial transaction real-time anti-fraud detection system according to claim 4, characterized in that: The risk warning module generates warning information containing detailed information according to the event type, and then sends it to the participants through multiple notification channels, tracks and records the response to the alarm, and regularly evaluates the effectiveness of the warning for subsequent analysis and improvement.

6. A digital financial transaction real-time anti-fraud detection system according to claim 5, further comprising: The transaction data collection module and the feedback learning module are characterized in that the transaction data collection module obtains digital financial transaction behavior data in real time, integrates the data and stores it in the system database, and the feedback learning module continuously adjusts and optimizes the model according to actual detection results and user feedback to improve detection accuracy and system adaptability.

7. A digital financial transaction real-time anti-fraud detection system according to claim 6, characterized in that: The transaction data collection module uses the AES encryption algorithm in GCM mode to encrypt the digital financial transaction behavior data, and interacts with the database through the LabSQL toolkit using Microsoft ADO technology and SQL language.

8. A real-time anti-fraud detection system for digital financial transactions according to claim 7, characterized in that: The feedback learning module uses user feedback as new training data, learns the characteristics of the new data through an online learning algorithm to update the detection model, and uses One-Class SVM anomaly detection to detect new and unseen fraud patterns.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the processor implements a real-time anti-fraud detection system for digital financial transactions as described in claims 1-8.

10. A terminal device, characterized in that: A real-time anti-fraud detection system for digital financial transactions according to any one of claims 1 to 8 is provided, comprising a processor and a computer-readable storage medium, the processor being used to implement various instructions; the computer-readable storage medium being used to store multiple instructions, the instructions being suitable for being loaded and executed by the processor.

Citation Information

Cited By

  • Payment fraud real-time identification and disposal method and system based on machine learning

    CN120996814A