Dynamic, graph-based system for anomaly detection for real-time risk assessment of maps

Abstract

A dynamic, graph-based anomaly detection system (100) for real-time map risk assessment, comprising: a data acquisition module configured to receive transaction data from one or more transaction sources in real time; a graph construction module configured to generate a dynamic graph comprising a variety of nodes representing entities such as cards, merchants, devices, accounts, and locations, as well as edges representing relationships between the entities; a graph update module configured to continuously update the dynamic graph based on incoming transaction data; an anomaly detection engine configured to analyze the dynamic graph to identify anomalous patterns that may indicate fraudulent activity; a feature extraction module configured to derive graph-based features from the dynamic graph; a machine learning module configured to process the graph-based features and generate a risk assessment associated with a transaction; and a real-time rating module configured to assign a risk rating to the transaction to enable fraud detection decisions.

Description

[0001] The present invention relates generally to the technical field of computer-implemented systems for the security of electronic transactions and in particular to a dynamic, graph-based anomaly detection system for the real-time risk assessment of cards.

[0002] The rapid growth of digital payment systems, including credit cards, debit cards, and online transaction platforms, has significantly increased the volume and speed of financial transactions. While such advances have improved convenience and accessibility, they have also led to a substantial rise in fraudulent activity, particularly no-physical-card transactions, identity theft, and account takeovers. Traditional fraud detection systems primarily rely on rule-based engines and static risk assessment models, using predefined rules and thresholds to flag suspicious transactions. However, such systems have several limitations.They are often unable to adapt to changing fraud patterns, require frequent manual updates, and tend to generate many false positives, impacting legitimate customer transactions and the overall user experience. In recent years, machine learning-based approaches have been introduced to improve the accuracy of fraud detection. These approaches typically analyze transaction attributes such as transaction amount, location, time, and merchant category. While these methods offer improvements over rule-based systems, they largely treat transactions as isolated events and fail to capture the complex relationships and interactions between entities such as cards, merchants, devices, and user behavior. In practice, fraudulent activity often exhibits interconnected and coordinated patterns, with multiple entities linked through hidden or indirect relationships.Existing systems lack the ability to dynamically model and analyze such interconnected structures in real time. Furthermore, traditional systems are ill-suited to handling highly dynamic and streaming data environments where transactional data must be processed with minimal latency to enable immediate risk assessment and decision-making. Graph-based approaches have been explored to represent relationships between entities; however, many existing solutions rely on static or periodically updated graphs, limiting their effectiveness in detecting real-time anomalies. Moreover, current implementations often face challenges regarding scalability, computational efficiency, and integration with real-time transaction processing systems.Accordingly, there is a need for an improved system and method that can dynamically create and update graphical representations of transaction ecosystems, efficiently detect anomalous patterns based on evolving relationships, and generate accurate, low-latency, real-time card risk assessments. The present invention addresses these challenges by providing a dynamic, graph-based anomaly detection system optimized for real-time fraud detection in electronic payment networks.

[0003] To solve this problem, the present invention provides a dynamic, graph-based anomaly detection system for real-time map risk assessment.

[0004] The system offers dynamic, graph-based anomaly detection to enable real-time card risk assessment in electronic payment networks.

[0005] The system dynamically creates and continuously updates graph structures that represent relationships between transaction entities such as cards, merchants, devices, and users.

[0006] The system identifies anomalous patterns by analyzing interconnected transaction histories, rather than treating transactions as isolated events.

[0007] The system uses graph-based machine learning models to improve the detection of complex and coordinated fraudulent activities.

[0008] The system performs processing of high-volume, low-latency streaming transaction data to enable immediate risk assessment.

[0009] The system generates adaptive and continuously updated risk assessments based on evolving transaction patterns and entity relationships.

[0010] The system reduces false alarms while ensuring high accuracy in fraud detection.

[0011] The system ensures scalability to process large-scale transaction networks in real time.

[0012] The system enables seamless integration into existing infrastructures for payment processing and fraud monitoring.

[0013] The system increases the security and reliability of electronic transactions by enabling the proactive detection and prevention of fraudulent activities.

[0014] The present invention provides a dynamic, graph-based anomaly detection system configured to perform real-time risk assessment for cards in electronic payment networks. The system comprises a data acquisition module configured to receive high-speed transaction data from multiple sources, including payment gateways, banking systems, and user devices. The system further comprises a graph construction module configured to model the received transaction data as a dynamic graph, with entities such as cards, merchants, devices, accounts, and locations represented as nodes and relationships between these entities represented as edges.The system further includes a graph update module configured to continuously update the dynamic graph in real time based on incoming transaction flows, thereby maintaining an up-to-date representation of the transaction ecosystem. The system also includes an anomaly detection engine configured to analyze the dynamic graph using graph-based learning techniques to identify suspicious patterns, abnormal connection formations, and deviations from established behavioral profiles. Finally, the system includes a feature extraction module configured to derive graph-based features, including node centrality, connectivity patterns, transaction frequency, and behavioral signatures associated with entities within the graph.The system also includes a machine learning module configured to process the extracted features and generate adaptive risk assessments for individual transactions or entities. Furthermore, the system includes a real-time scoring module configured to assign a risk score to each transaction with minimal latency, enabling immediate decision-making, including approval, rejection, or further review of the transaction. The system also includes an alert generation module configured to trigger alerts for transactions that exceed predefined or dynamically adjusted risk thresholds. In one embodiment, the system utilizes graphical neural networks or similar advanced analytical models to improve detection accuracy in coordinated and multi-entity fraud scenarios.In another embodiment, the system includes adaptive learning mechanisms to continuously refine recognition models based on historical data and feedback. Advantageously, the system enables improved detection of complex fraud patterns, reduces false alarms, and ensures scalable and efficient processing of large volumes of transaction data in real time. The invention thus improves the overall security, reliability, and efficiency of electronic payment systems.

[0015] Fig. The system (100) for dynamic, graph-based anomaly detection is shown, which is designed to perform a map risk assessment in real time.

[0016] Fig.Figure 100 illustrates the dynamic, graph-based anomaly detection system configured to perform real-time card risk assessment. The system includes a data acquisition module that receives transaction data and forwards it to a graph construction module and a graph update module. The graph construction module is configured to generate a dynamic graph representing entities as nodes and relationships as edges, while the graph update module is configured to continuously update the dynamic graph based on incoming transaction streams. The dynamic graph serves as the central processing structure connected to downstream analysis components.The system further includes a feature extraction module and an anomaly detection engine, functionally coupled to the dynamic graph. The feature extraction module derives graph-based features, and the anomaly detection engine identifies anomalous patterns within the graph. The outputs of both the feature extraction module and the anomaly detection engine are provided to a machine learning module configured to generate predictive insights. The system also includes a real-time risk assessment (scoring) module configured to receive the outputs from the machine learning module and assign risk scores to transactions, enabling real-time fraud detection and decision-making.

[0017] The present invention provides the system (100) for dynamic, graph-based anomaly detection, configured to perform real-time card risk assessment in electronic payment networks. The system (100) comprises a data acquisition module configured to receive high-speed transaction data from multiple sources, including payment gateways, banking systems, merchant platforms, and user devices, and further configured to preprocess the received data by normalization and validation. The system (100) also comprises a graph construction module configured to construct a dynamic graph in which entities such as cards, merchants, devices, accounts, and locations are represented as nodes, and relationships between such entities are represented as edges associated with attributes such as timestamps, transaction amounts, and interaction frequency.The system (100) also includes a graph update module configured to continuously update the dynamic graph in real time by incrementally adding or modifying nodes and edges based on incoming transaction data without recalculating the entire graph structure, thus ensuring computational efficiency and scalability in high-transaction-volume environments.

[0018] The system (100) further comprises an anomaly detection engine configured to analyze the dynamically evolving graph to identify anomalous patterns indicative of fraudulent activity by detecting deviations in connectivity, behavioral patterns, and the formation of suspicious clusters among interconnected entities. The system (100) also includes a feature extraction module configured to derive graph-based features such as node centrality, clustering coefficients, degree distribution, and temporal interaction features, and a machine learning module configured to process the extracted features using graph-based learning models, including graphical neural networks, to generate predictive risk insights.The system (100) further includes a real-time assessment module configured to assign a risk rating to each transaction with low latency, enabling immediate decision-making including approval, rejection, or further review, and an alert generation module configured to generate alerts when the risk rating exceeds predefined or dynamically set thresholds. The system (100) thus enables the accurate detection of complex and coordinated fraud patterns, reduces false positives, ensures scalability, and improves the security and reliability of electronic transaction systems through proactive and adaptive fraud detection.

Claims

[1] A dynamic, graph-based anomaly detection system (100) for real-time map risk assessment, comprising: a data acquisition module configured to receive transaction data from one or more transaction sources in real time; a graph construction module configured to generate a dynamic graph comprising a variety of nodes representing entities such as cards, merchants, devices, accounts, and locations, as well as edges representing relationships between the entities; a graph update module configured to continuously update the dynamic graph based on incoming transaction data; an anomaly detection engine configured to analyze the dynamic graph to identify anomalous patterns that may indicate fraudulent activity; a feature extraction module configured to derive graph-based features from the dynamic graph; a machine learning module configured to process the graph-based features and generate a risk assessment associated with a transaction; and a real-time rating module configured to assign a risk rating to the transaction to enable fraud detection decisions. [2] System (100) according to claim 1, wherein the data acquisition module is configured to process high-speed streaming transaction data from multiple heterogeneous sources. [3] System (100) according to claim 1, wherein the graph construction module is configured to represent temporal relationships between entities using time-stamped edges. [4] System (100) according to claim 1, wherein the graph update module is configured to incrementally update the graph without recalculating the entire graph structure. [5] System (100) according to claim 1, wherein the anomaly detection engine is configured to detect deviations based on historical behavior patterns of entities. [6] System (100) according to claim 1, wherein the feature extraction module is configured to calculate features including node centrality, clustering coefficients, connectivity patterns and transaction frequency. [7] System (100) according to claim 1, wherein the machine learning module comprises one or more graph-based learning models, including graphical neural networks. [8] System (100) according to claim 1, wherein the machine learning module is configured to adaptively update model parameters based on feedback from previously processed transactions. [9] System (100) according to claim 1, further comprising a module for generating warning messages, which is configured to generate warning messages when the risk value exceeds a predefined or dynamically determined threshold.

Images