An adaptive label-aware graph neural network credit card fraud detection method
By constructing a multi-relationship heterogeneous graph and processing node features of different neighbor types in the graph aggregation layer, the shortcomings of existing models in distinguishing neighbor differences are addressed, achieving higher accuracy and robustness in credit card fraud detection.
Patent Information
- Application Number
- CN202510055244.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-14
- Publication Date
- 2026-02-17
- Estimated Expiration
- 2045-01-14
AI Technical Summary
Existing graph neural network-based credit card fraud detection models do not effectively distinguish and handle differences when dealing with neighbors of different label types, affecting the accuracy and robustness of fraud detection.
An adaptive label-aware graph neural network approach is adopted. By constructing a multi-relationship heterogeneous graph, a multi-layer graph aggregation layer is used to process node features, including heterogeneous graph decomposition, attention module, transformation module and multi-head attention aggregation module, to handle fraudulent, good and unknown neighbor types respectively, and the model parameters are optimized by loss function.
It improves the accuracy and robustness of credit card fraud detection, better captures complex transaction patterns and high-level interactions between entities, and enhances the model's performance in complex fraud detection tasks.
Smart Images
Figure CN119991129B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of financial transaction fraud detection technology, and relates to an adaptive label-aware graph neural network method for credit card fraud detection. Background Technology
[0002] Credit card fraud detection is a key challenge in the field of financial security. With the widespread adoption of electronic payments, fraudulent activities have become increasingly complex and difficult to identify. Traditional methods, such as rule-based systems and machine learning algorithms, while capable of identifying and detecting fraudulent activities to some extent, face numerous challenges when dealing with large-scale, high-dimensional, complex, and ever-changing transaction data.
[0003] With the development of deep learning, especially the introduction of Graph Neural Networks (GNNs), new solutions have been provided for credit card fraud detection. Credit card transaction data can typically be represented by a graph structure, where entities such as users, merchants, and transactions can be modeled as nodes in the graph, and the transaction behaviors between these nodes can be viewed as edges. GNNs excel at handling this non-Euclidean data structure, capturing complex transaction patterns and high-order interactions between entities by modeling nodes and edges, which aligns perfectly with the inherent structure of credit card transaction data. Among these, Graph Convolutional Networks (GCNs) extend convolution operations to non-Euclidean data, effectively aggregating information from the node's neighborhood to capture the local structural features of the graph. Another approach, Graph Attention Networks (GATs), enhances the model's expressive power and flexibility during information aggregation by assigning different importance weights to different neighboring nodes. GAT's multi-head attention mechanism not only improves model stability but also allows for the capture of multi-scale node interactions, further enhancing the model's performance in complex fraud detection tasks.
[0004] Current credit card fraud detection methods have shortcomings: existing graph neural network (GNN)-based fraud detection models do not effectively distinguish and handle differences in neighbor types with different labels, thus affecting the accuracy and robustness of fraud detection. Summary of the Invention
[0005] To address the problems of the prior art, this invention employs an adaptive label-aware graph neural network method for credit card fraud detection, comprising: acquiring credit card transaction data, inputting the credit card transaction data into a trained credit card fraud detection model, and obtaining credit card fraud detection results; the credit card fraud detection model comprises: a graph neural network and a classification module; the graph neural network comprises multiple graph aggregation layers connected in series.
[0006] The training process for a credit card fraud detection model includes:
[0007] S1: Obtain the credit card transaction record dataset and construct an initial multi-relation heterogeneous graph based on the dataset; where the nodes of the initial multi-relation heterogeneous graph are credit card transaction records, and the edges are the relationships between the credit card transaction records;
[0008] S2: Input the initial multi-relation heterogeneous graph into a series of interconnected graph aggregation layers, and the last graph aggregation layer outputs the updated multi-relation heterogeneous graph and its node features;
[0009] S3: Input the node features of the multi-relationship heterogeneous graph output from the last graph aggregation layer into the classification module to obtain the fraud detection results;
[0010] S4: Calculate the loss function value based on the fraud detection results, update the model parameters based on the loss function value, and obtain the trained credit card fraud detection model when the loss function value is minimized.
[0011] The credit card transaction record dataset includes multiple transaction records between users and credit card providers. Each transaction record includes a real label and multiple attribute features. Constructing an initial multi-relationship heterogeneous graph based on the credit card transaction record dataset involves: treating transaction records as nodes, classifying nodes into fraud nodes, good nodes, and unknown nodes based on the real label of the transaction record; treating the attribute features of the transaction records as node features, selecting multiple attribute features from the attribute features of the transaction records, using the selected attribute features as the relation types of the edges, and constructing edges corresponding to the relation types between transaction record nodes based on the selected attribute features.
[0012] The sequentially cascaded multi-layer graph aggregation process for the initial multi-relationship heterogeneous graph includes:
[0013] S21: Input the initial multi-relation heterogeneous graph into the first graph aggregation layer to obtain the multi-relation heterogeneous graph output by the first graph aggregation layer;
[0014] S22: Input the multi-relation heterogeneous graph output by the first graph aggregation layer into the second graph aggregation layer to obtain the multi-relation heterogeneous graph output by the second graph aggregation layer;
[0015] S23: Input the multi-relationship heterogeneous graph output from the previous graph aggregation layer into the current graph aggregation layer to obtain the multi-relationship heterogeneous graph output from the current graph aggregation layer;
[0016] S24: Repeat step S23 until the multi-relation heterogeneous graph output by the last graph aggregation layer is obtained.
[0017] Each graph aggregation layer includes: a heterogeneous graph decomposition module, an attention module, a transformation module, and a multi-head attention aggregation module; the current graph aggregation layer l processes the multi-relationship heterogeneous graph output by the previous graph aggregation layer l-1, including:
[0018] S231. Input the multi-relationship heterogeneous graph output from the aggregation layer l-1 of the previous graph into the heterogeneous graph decomposition module to obtain R subgraphs.
[0019] S232. Each subgraph corresponds to an attention module. Input the R subgraphs into their respective attention modules to obtain the node features of the R subgraphs under each neighbor type k. Where r is the index of the subgraph and i is the index of the node;
[0020] S233. Each subgraph corresponds to a transformation module, which transforms the node features of the R subgraphs under each neighbor type. Node features of the multi-relation heterogeneous graph output by aggregation layer l-1 in the previous figure Input the corresponding transformation module to obtain the final node features of R subgraphs.
[0021] S234. Analyze the final node features of the R subgraphs. The input is aggregated by the multi-head attention aggregation module to obtain the multi-relation heterogeneous graph and its node features output by the current graph aggregation layer l.
[0022] The attention module processes the corresponding subgraph in the following ways:
[0023]
[0024] in, Let σ represent the features of node i in subgraph r under the k-th type of neighbors in the attention module of the current graph aggregation layer l, where σ is the activation function and j is the k-th type of neighbor node of node i in subgraph r. Let be the set of the k-th type of neighboring nodes of node i in subgraph r. Let i be the attention coefficients of nodes i and j in subgraph r within the attention module of the current graph aggregation layer l. Let be the weight matrix of the subgraph r for the k-th type of neighbor nodes. The features of node j in the multi-relation heterogeneous graph output by the (l-1)th layer graph aggregation layer.
[0025] The conversion module includes: a self-conversion module, a fraudulent neighbor type conversion module, a benign neighbor type conversion module, and an unknown neighbor type conversion module. The conversion module processes the node features of the corresponding subgraph r under each neighbor type, including: [processing the node features of subgraph r under each neighbor type]. Input the corresponding fraudulent neighbor type conversion module, benign neighbor type conversion module, and unknown neighbor type conversion module respectively; then input the node features. The input is the self-transformation module. The outputs of the self-transformation module, the fraudulent neighbor type transformation module, the benign neighbor type transformation module, and the unknown neighbor type transformation module are combined to obtain the final node features of R subgraphs.
[0026] The weight matrices for the fraudulent neighbor type conversion module, the benign neighbor type conversion module, and the unknown neighbor type conversion module are as follows: as well as in, Adjust the weights based on node type.
[0027] weight matrix Node type adjustment weight Among them, Ψ fr Ψ be Both Φ and Φ are single-layer perceptrons.
[0028] The multi-head attention aggregation module includes multiple attention heads. The aggregation of the final node features of the R subgraphs by the multi-head attention aggregation module involves: inputting the final node features of the R subgraphs into each attention head, concatenating the outputs of each attention head, and obtaining the features of each node i in the multi-relation heterogeneous graph output by the current graph aggregation layer l.
[0029] Each attention head aggregates the final node features of R subgraphs, including:
[0030]
[0031] in, Let σ be the feature of node i in the attention head q, σ be the activation function, and r be the index of the subgraph. Let i be the attention coefficient of node i in subgraph r within attention head q. The weight matrix of subgraph r in attention head q.
[0032] loss function for:
[0033]
[0034] in, y represents the set of nodes in a preliminary multi-relation heterogeneous graph. i For the true label of node i, This represents the probability that node i is a fraudulent node.
[0035] Beneficial effects:
[0036] 1. This invention constructs a multi-relation heterogeneous graph by selecting multiple attribute features as relation types from a credit card transaction record dataset, and updates node features under each relation type in the graph aggregation layer. This fully utilizes the neighbor information of node features in multiple relation types, improving the model's ability to handle complex credit card fraud detection tasks. 2. In the graph aggregation layer, this invention uses an attention mechanism to aggregate node features for each relation, and further processes different types of neighbors using transformation functions. This allows nodes to adaptively adjust the information aggregated from their heterogeneous and homogeneous neighbors, improving the ability to detect credit card fraud. Attached Figure Description
[0037] Figure 1 This is a structural diagram of the credit card fraud detection model provided in an embodiment of the present invention;
[0038] Figure 2 This is a schematic diagram illustrating the connection between the central node and its different types of neighboring nodes provided in an embodiment of the present invention. Detailed Implementation
[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0040] like Figure 1 , Figure 2 As shown, this embodiment of the invention employs an adaptive label-aware graph neural network method for credit card fraud detection, comprising: acquiring credit card transaction data, inputting the credit card transaction data into a trained credit card fraud detection model, and obtaining credit card fraud detection results; the credit card fraud detection model includes: a graph neural network and a classification module; the graph neural network includes multiple graph aggregation layers connected in series.
[0041] The training process for a credit card fraud detection model includes:
[0042] S1: Obtain the credit card transaction record dataset and construct an initial multi-relation heterogeneous graph based on the dataset; where the nodes of the initial multi-relation heterogeneous graph are credit card transaction records, and the edges are the relationships between the credit card transaction records;
[0043] The credit card transaction record dataset includes multiple actual transaction records from users and credit card providers. Each transaction record includes a real label and multiple attribute features. The labels include fraud (fr), good (be), and unknown (un), and the attribute features include transaction type, transaction location (city), transaction amount, transaction time, etc. Multiple attribute features are selected from the attribute features of the transaction records, and each selected attribute feature is used as a relation type. A multi-relation heterogeneous graph is constructed based on the similarity of the transaction records on each selected attribute feature. The nodes of the multi-relation heterogeneous graph are transaction records, which are classified into fraud nodes, good nodes, and unknown nodes according to the real label of the transaction records. The edges represent the relationships between transaction records on each selected attribute feature.
[0044] In one embodiment, a partial data example for constructing a multi-relationship heterogeneous graph is shown in Table 1. It can be observed that transactions time0, time2, time4, and time5 have the same type, so the following connections are made based on transaction type: time0->time2, time4->time0, time5->time2, time4->time5, time5->time0, time2->time4. Similarly, transactions time0, time2, time3, time4, and time5 have the same address, so the following connections are made based on transaction address: time2->time0, time3->time0, time2->time4, time3->time5, time4->time5, time0->time4, time0->time5, time2->time3, time4->time3, time5->time2. The number of connections is selected by setting a connection limit for a single point and single relationship and limiting the maximum time span (e.g., in transaction data, the maximum number of neighbors connected by attribute features is set to 8, while the time span of the connected transaction records is limited to 500 time slices).
[0045] Table 1
[0046]
[0047]
[0048] S2: Input the initial multi-relation heterogeneous graph into a series of interconnected graph aggregation layers, and the last graph aggregation layer outputs the updated multi-relation heterogeneous graph and its node features;
[0049] The sequentially cascaded multi-layer graph aggregation process for the initial multi-relationship heterogeneous graph includes:
[0050] S21: Input the initial multi-relation heterogeneous graph into the first graph aggregation layer to obtain the multi-relation heterogeneous graph output by the first graph aggregation layer;
[0051] S22: Input the multi-relation heterogeneous graph output by the first graph aggregation layer into the second graph aggregation layer to obtain the multi-relation heterogeneous graph output by the second graph aggregation layer;
[0052] S23: Input the multi-relationship heterogeneous graph output from the previous graph aggregation layer into the current graph aggregation layer to obtain the multi-relationship heterogeneous graph output from the current graph aggregation layer;
[0053] S24: Repeat step S23 until the multi-relation heterogeneous graph output by the last graph aggregation layer is obtained.
[0054] Each graph aggregation layer includes: a heterogeneous graph decomposition module, an attention module, a transformation module, and a multi-head attention aggregation module; the current graph aggregation layer l processes the multi-relationship heterogeneous graph output by the previous graph aggregation layer l-1, including:
[0055] S231. Input the multi-relationship heterogeneous graph output from the aggregation layer l-1 of the previous graph into the heterogeneous graph decomposition module to obtain R subgraphs.
[0056] S232. Each subgraph corresponds to an attention module. Input the R subgraphs into the corresponding attention modules to obtain the node features of the R subgraphs under each neighbor type.
[0057] The attention module processes the corresponding subgraph r in the following ways:
[0058]
[0059] in, Let σ represent the features of node i in subgraph r under the k-th type of neighbors in the attention module of the current graph aggregation layer l, where σ is the activation function and j is the k-th type of neighbor node of node i in subgraph r. Let be the set of the k-th type of neighboring nodes of node i in subgraph r. Let i be the attention coefficients of nodes i and j in subgraph r within the attention module of the current graph aggregation layer l. Let be the weight matrix of the subgraph r for the k-th type of neighbor nodes. The features of node j in the multi-relation heterogeneous graph output by the (l-1)th layer graph aggregation layer.
[0060] Attention coefficient The calculation formula is:
[0061]
[0062] Here, || represents the concatenation operation. Let j' be a learnable vector used to measure the importance of node pairs, where j' is a neighboring node of node i in subgraph r. Let be the weight matrix of the attention head k of the graph attention aggregation layer corresponding to subgraph r in the l-th graph aggregation layer.
[0063] S233. Each subgraph corresponds to a transformation module, which combines the node features of the R subgraphs under each neighbor type with the node features of the multi-relationship heterogeneous graph output by the aggregation layer l-1 of the previous graph. Input the corresponding transformation module to obtain the final node features of R subgraphs;
[0064] The conversion module includes: a self-conversion module, a fraudulent neighbor type conversion module, a benign neighbor type conversion module, and an unknown neighbor type conversion module; the conversion module processes the node characteristics of the corresponding subgraph r under each neighbor type, including:
[0065] The node features of subgraph r under each neighbor type are input into the corresponding fraudulent neighbor type conversion module, benign neighbor type conversion module, and unknown neighbor type conversion module for further conversion and processing. This further refines the contribution of different types of neighbors, thereby effectively improving the quality of node representation and detection performance. The features of node i in subgraph r output by the conversion module at layer l are... as follows:
[0066]
[0067] Among them, Comb (l) For combined operations, and All are MLPs (Multilayer Perceptrons), and their respective transformation functions are: self-transformation function, fraudulent neighbor type transformation function, benign neighbor type transformation function, and unknown neighbor type transformation function. Each transformation function is defined by a corresponding weight matrix. Parameterization.
[0068] To better reflect the uncertainty of unknown neighbors, the weight matrix of the fraudulent neighbor type conversion function and the benign neighbor type conversion function is used as the weight matrix of the unknown neighbor type conversion function:
[0069]
[0070] in, It is a scalar used to adjust the category trend of unknown neighbors associated with the central node i.
[0071] It will adaptively adjust based on the characteristics of the central node i, defined as a function of node i: It is a single-layer MLP with sigmoid activation shared across all nodes.
[0072] weight matrix and Specifically for the central node i, it will adaptively adjust according to the characteristics of the central node i, defined as:
[0073]
[0074] Among them, Ψ ft Ψ be : Implemented using a single-layer MLP:
[0075]
[0076] Where, d ′ Let be the dimension of the node features, and diag() be the diagonal matrix of the input features.
[0077] S234. Input the final node features of the R subgraphs into the multi-head attention aggregation module for aggregation to obtain the multi-relationship heterogeneous graph output by the current graph aggregation layer.
[0078] Specifically, each attention head aggregates the final node features of the R subgraphs as follows:
[0079]
[0080] in, Let σ be the feature of node i in the attention head q, σ be the activation function, and r be the index of the subgraph.
[0081] Let be the attention coefficients of the nodes in subgraph r within attention head q. Let N be the weight matrix of subgraph r in attention head q. a For the number of heads.
[0082] Attention coefficient The calculation is as follows:
[0083]
[0084] Where || denotes the concatenation operation, and σ is the activation function. This is a learnable vector used to measure the importance of relation r. Let R be the weight matrix of the attention head q, and R be the number of relations.
[0085] The features output by each attention head are concatenated to obtain the features of each node i in the multi-relation heterogeneous graph output by the current graph aggregation layer l.
[0086] S3: Input the node features of the multi-relationship heterogeneous graph output from the last graph aggregation layer into the classification module to obtain the fraud detection results;
[0087] Specifically, after multiple iterations, the final node representation is input into a multilayer perceptron (MPL) for fraud detection.
[0088]
[0089] in, This represents the probability that node i is a fraudulent node. This is the output of the Lth graph aggregation layer.
[0090] S4: Calculate the loss function value based on the fraud detection results, update the model parameters based on the loss function value, and obtain the trained credit card fraud detection model when the loss function value is minimized.
[0091] loss function Defined as:
[0092]
[0093] in, y represents the set of nodes in a multi-relation heterogeneous graph. i Let i be the actual label of node i.
[0094] The above-described embodiments further illustrate the purpose, technical solution, and advantages of the present invention. It should be understood that the above-described embodiments are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made to the present invention within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. An adaptive label-aware graph neural network method for credit card fraud detection, characterized in that, include: Obtain credit card transaction data, input the credit card transaction data into the trained credit card fraud detection model, and obtain the credit card fraud detection results; The credit card fraud detection model includes a graph neural network and a classification module; the graph neural network consists of multiple graph aggregation layers connected in series. The training process for a credit card fraud detection model includes: S1: Obtain the credit card transaction record dataset and construct an initial multi-relation heterogeneous graph based on the dataset; where the nodes of the initial multi-relation heterogeneous graph are credit card transaction records, and the edges are the relationships between the credit card transaction records; The credit card transaction record dataset includes multiple transaction records between users and credit card providers. Each transaction record includes a real label and multiple attribute features. Constructing an initial multi-relationship heterogeneous graph based on the credit card transaction record dataset involves: treating transaction records as nodes, classifying nodes into fraud nodes, good nodes, and unknown nodes based on the real label of the transaction record; treating the attribute features of the transaction records as node features, selecting multiple attribute features from the attribute features of the transaction records, using the selected attribute features as the relation types of the edges, and constructing edges corresponding to the relation types between transaction record nodes based on the selected attribute features. S2: Input the initial multi-relation heterogeneous graph into a series of interconnected graph aggregation layers to obtain the multi-relation heterogeneous graph and its node features output by the last graph aggregation layer; The sequentially cascaded multi-layer graph aggregation process for the initial multi-relationship heterogeneous graph includes: S21: Input the initial multi-relation heterogeneous graph into the first graph aggregation layer to obtain the multi-relation heterogeneous graph output by the first graph aggregation layer; S22: Input the multi-relation heterogeneous graph output by the first graph aggregation layer into the second graph aggregation layer to obtain the multi-relation heterogeneous graph output by the second graph aggregation layer; S23: Input the multi-relationship heterogeneous graph output from the previous graph aggregation layer into the current graph aggregation layer to obtain the multi-relationship heterogeneous graph output from the current graph aggregation layer; S24: Repeat step S23 until the multi-relation heterogeneous graph output by the last graph aggregation layer is obtained; Each graph aggregation layer includes: a heterogeneous graph decomposition module, an attention module, a transformation module, and a multi-head attention aggregation module; the current graph aggregation layer l processes the multi-relationship heterogeneous graph output by the previous graph aggregation layer l-1, including: S231. Input the multi-relationship heterogeneous graph output from the aggregation layer l-1 of the previous graph into the heterogeneous graph decomposition module to obtain R subgraphs. S232. Each subgraph corresponds to an attention module. Input the R subgraphs into their respective attention modules to obtain the node features of the R subgraphs under each neighbor type k. Where r is the index of the subgraph and i is the index of the node; S233. Each subgraph corresponds to a transformation module, which transforms the node features of the R subgraphs under each neighbor type. The node features of the multi-relation heterogeneous graph output by the aggregation layer l-1 in the previous figure. Input the corresponding transformation module to obtain the final node features of R subgraphs. ; S234. Analyze the final node features of the R subgraphs. The input is aggregated by the multi-head attention aggregation module to obtain the multi-relationship heterogeneous graph and its node features output by the current graph aggregation layer l. ; The conversion module includes: a self-conversion module, a fraudulent neighbor type conversion module, a benign neighbor type conversion module, and an unknown neighbor type conversion module. The conversion module processes the node features of the corresponding subgraph r under each neighbor type, including: [processing the node features of subgraph r under each neighbor type]. Input the corresponding fraudulent neighbor type conversion module, benign neighbor type conversion module, and unknown neighbor type conversion module respectively; then input the node features. The input is the self-transformation module. The outputs of the self-transformation module, the fraudulent neighbor type transformation module, the benign neighbor type transformation module, and the unknown neighbor type transformation module are combined to obtain the final node features of R subgraphs. ; S3: Input the node features of the multi-relationship heterogeneous graph output from the last graph aggregation layer into the classification module to obtain the fraud detection results; S4: Calculate the loss function value based on the fraud detection results, update the model parameters based on the loss function value, and obtain the trained credit card fraud detection model when the loss function value is minimized.
2. The adaptive label-aware graph neural network method for credit card fraud detection according to claim 1, characterized in that, The attention module processes the corresponding subgraph in the following ways: in, Let σ be the feature of node i in subgraph r under the k-th type of neighbor in the attention module of the current graph aggregation layer l, and let σ be the activation function. Let k be the neighbor node of the kth type of node i in subgraph r. Let be the set of the k-th type of neighboring nodes of node i in subgraph r. Let i be the attention coefficients of nodes i and j in subgraph r within the attention module of the current graph aggregation layer l. Let be the weight matrix of the subgraph r for the k-th type of neighbor nodes. The nodes of the multi-relation heterogeneous graph output by the (l-1)th layer graph aggregation layer Its characteristics.
3. The adaptive label-aware graph neural network method for credit card fraud detection according to claim 1, characterized in that, The weight matrices for the fraudulent neighbor type conversion module, the benign neighbor type conversion module, and the unknown neighbor type conversion module are as follows: , as well as ;in, , Adjust the weights based on node type.
4. The adaptive label-aware graph neural network method for credit card fraud detection according to claim 3, characterized in that, weight matrix , Adjusting weights ;in, All of them are single-layer perceptrons.
5. The adaptive label-aware graph neural network method for credit card fraud detection according to claim 1, characterized in that, The multi-head attention aggregation module includes multiple attention heads; The multi-head attention aggregation module aggregates the final node features of R subgraphs by: inputting the final node features of the R subgraphs into each attention head, and concatenating the outputs of each attention head to obtain the current graph aggregation layer. Features of each node i in the output multi-relation heterogeneous graph ; Each attention head aggregates the final node features of R subgraphs, including: in, Let σ be the feature of node i in the attention head q, σ be the activation function, and r be the index of the subgraph. Let i be the attention coefficient of node i in subgraph r within attention head q. The weight matrix of subgraph r in attention head q.
6. The adaptive label-aware graph neural network method for credit card fraud detection according to claim 1, characterized in that, loss function for: in, This represents the set of nodes in a preliminary multi-relation heterogeneous graph. For nodes The true label, Represents a node The probability of a node being fraudulent.