CN110084609B - Transaction fraud behavior deep detection method based on characterization learning - Google Patents
Transaction fraud behavior deep detection method based on characterization learning Download PDFInfo
- Publication number
- CN110084609B CN110084609B CN201910327470.1A CN201910327470A CN110084609B CN 110084609 B CN110084609 B CN 110084609B CN 201910327470 A CN201910327470 A CN 201910327470A CN 110084609 B CN110084609 B CN 110084609B
- Authority
- CN
- China
- Prior art keywords
- data
- transaction
- data set
- model
- fraud
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q20/00—Payment architectures, schemes or protocols
- G06Q20/38—Payment protocols; Details thereof
- G06Q20/40—Authorisation, e.g. identification of payer or payee, verification of customer or shop credentials; Review and approval of payers, e.g. check credit lines or negative lists
- G06Q20/401—Transaction verification
- G06Q20/4016—Transaction verification involving fraud or risk level assessment in transaction processing
Abstract
The invention relates to a transaction fraud depth detection method based on characterization learning, which is characterized by comprising the following steps: establishing a representation learning multi-presentation structure based on an Autoencoder model, and extracting a data set from transaction data by a bagingbalance method to obtain a corresponding representation learning vector; and establishing an OSD-DF classifying structure, taking the characterization learning vector obtained by the multi-presenting structure as the input of the OSD-DF classifying structure, training the OSD-DF classifying structure, and obtaining the fraud detection result of the transaction data. The electronic transaction fraud detection method provided by the invention can effectively detect fraud in electronic transactions. The method provided by the invention establishes a framework of the electronic transaction fraud detection method from the practical point of view, and provides technical support for solving the fraud transaction detection.
Description
Technical Field
The invention relates to a network transaction detection method, and belongs to the technical field of information.
Background
With the recent development of internet finance, electronic transaction fraud detection has become an popular research field including credit card fraud detection, mobile payment fraud detection, B2C (Business-to-Customer) transaction fraud detection, etc.
Fraud detection algorithms based on machine learning are very widely used in the field of electronic transaction fraud detection. The supervised learning model establishes a fraud detection model based on historical transaction data after manual verification of the tags, thereby determining whether a new transaction has a likelihood of fraud. In 2018, shiyang Xuan et al used two random forests of different base classifiers to train the behavioral characteristics of normal and abnormal transactions, respectively, and analyzed the performance of the two random forests on credit card fraud detection. In addition, unsupervised learning models typically employ outlier detection or anomaly detection techniques that treat the identified outliers as detected fraudulent transactions. In 2014, olszewski D built a user behavior model using self-organization map (SOM) method, finding outliers deviating from normal user behavior for fraud detection. Machine learning based detection algorithms can learn known patterns of fraud and detect potential new fraud strategies. However, the detection method with supervised learning is strongly dependent on the correctness of the original mark, and the existing sample imbalance problem needs to be dealt with. Unsupervised learning is very sensitive to overlapping distributions of normal and fraudulent transactions, often resulting in a serious drop in accuracy.
With the excellent performance of deep learning technology on numerous classification tasks, the field of electronic transaction fraud detection began to introduce deep learning technology. As a conclusion drawn by one article of the consultation company McKinsey, month 4 2017: deep learning can provide a solution to the financial fraud problem by providing real-time transaction details in all historical customer data and transaction records. In 2016, kang Fu et al proposed a convolutional neural network (Convolutional Neural Network, CNN) based fraud detection framework that learns the intrinsic pattern of fraud from tagged transaction data and exhibits excellent performance on experimental data of a mainstream commercial bank. In 2017, shuhao Wang et al, beijing dong finance, proposed a novel fraud detection system based on deep learning, CLUE. The system captures more transaction detail information mainly through an ebadd-based neural network and a cyclic neural network (Recurrent Neural Networks, RNN), thereby achieving effective fraud detection. In 2018, zhaohui Zhang et al proposed a CNN network transaction fraud detection model with a feature alignment layer. Through different convolution pattern learning, the model shows better performance than the existing CNN detection model on the real transaction data of a commercial bank. In the same year, abhimanyu Roy et al have studied the application of deep learning model in credit card fraud detection task, solve the ubiquitous problem in fraud with high-performance distributed cloud computing environment, and provide a parameter adjustment framework of deep learning topology to enable financial institutions to reduce losses by preventing fraudulent activities. However, feature learning in a single transaction is not sufficient for deep learning models, although more sequential information between transactions can be extracted by deep learning techniques. These relationships in a single transaction record may be well learned by some machine learning model (e.g., RF), but at the cost of impaired time-series learning capabilities.
Thus, research in recent years into electronic transaction fraud detection has begun to emerge as detection techniques that combine the respective advantages of machine learning and deep learning. In 2017, xurui Li et al proposed a new sandwich-like time-series learning model of "witin-betwen-witin" (WBW). By combining the integrated learning model and the deep learning model and simultaneously introducing an attention mechanism, the effect of fraud detection is further improved. In the same year, zahra Kazemi and Houman Zarrabi use an automatic encoder (autoencoder) to provide optimal feature information in credit card transaction records, and then add a softmax neural network as a classifier to determine if the transaction is fraudulent. Experiments reveal the advantages of the proposed fraud detection model over other methods.
Under the big trend of internet finance, digitization technologies such as big data, artificial intelligence and the like are widely applied to the finance field, and the volume and development potential of the finance market are gradually amplified. At the same time, the risk hidden trouble of exposure is increased, and the fraud phenomenon is endangered. According to statistics, the Chinese fraud practitioner in 2017 has more than 150 ten thousand, annual output value reaches the trillion level, a financial institution applying the internet financial technology to develop financial services is one of main subjects of attack, and the wind control link of digital finance faces a great challenge.
In this case, accurate detection of fraudulent transaction behavior is a very important research direction in the field of electronic transaction fraud detection. The traditional expert rule-driven fraud detection technology requires a large amount of manual operation, has high application cost and low efficiency, and meanwhile, the traditional anti-fraud means has single dimension, so that a multi-dimensional user portrait is difficult to form for a user. The electronic transaction has the characteristics of strong real-time performance, large data volume and small amount and high frequency of fraud, and the traditional anti-fraud means can hardly accurately detect the fraud of the electronic transaction.
At present, a great deal of research mainly based on machine learning is widely applied to the field of fraud detection, and comprises machine learning algorithms such as decision trees, support vector machines, naive Bayes, random forests and the like. Machine learning techniques accurately detect electronic transactions with the potential for fraud by learning historical transaction information for the electronic transactions, learning existing fraud strategies, and mining potential fraud strategies. In addition, some research that considers deep learning techniques is also increasingly being used in fraud detection tasks.
Deep learning techniques such as convolutional neural networks and recurrent neural networks have achieved excellent performance in a number of popular fields, such as image recognition, natural language processing, and the like. Deep learning techniques are excellent in handling high-dimensional data and nonlinear feature space input, which also has commonality problems in fraud detection tasks. Based on this, some researches begin to introduce deep learning technology for fraud detection, and the problem of insufficient feature expression in fraud detection is solved by utilizing the strong feature learning capability shown by deep learning. However, the problem of fraud detection is not completely solved using only machine learning techniques or deep learning techniques.
Disclosure of Invention
The purpose of the invention is that: the advantages of integrating machine learning and deep learning are used for fraud detection.
In order to achieve the above purpose, the technical scheme of the invention provides a transaction fraud depth detection method based on characterization learning, which is characterized by comprising the following steps:
(1) Establishing a representation learning multi-presentation structure based on an Autoencoder model, extracting a data set from transaction data through a bagging balance method, inputting the data set into the multi-presentation structure to obtain a corresponding representation learning vector, wherein the extracting the data set from the transaction data through the bagging balance method comprises the following steps of:
s301, dividing data set
Dividing transaction data into sample sets D of legal transactions major And sample set D of fraudulent transactions minor ；
S302, data downsampling
Sample set D is randomly sampled every time major Selecting a sample copy to be placed in the data set D sample Then put the sample back into sample set D major In which the sampled data volume is the sample set D minor Is of a size of (2);
s303, feature sampling
Randomly selecting a feature subset from the feature space, and performing feature filtering on the downsampled data;
s304, returning a data set;
(2) An OSD-DF classifying structure is established, a multi-presenting structure is used for obtaining a characterization learning vector as the input of the OSD-DF classifying structure, the OSD-DF classifying structure is trained, and a fraud detection result of transaction data is obtained, comprising the following steps:
s201, data input
Acquiring a characterization learning vector obtained by a multi-presentation structure as an input of an OSD-DF classification structure;
s202, initializing a model
Setting a model structure of an OSD-DF classifying structure, inputting data into a processing layer of the OSD-DF classifying structure, and outputting a processing result to a next layer;
s203, training model
The next layer receives the characteristic information processed by the previous layer, the operation of the step S202 is repeated, the next layer is continuously expanded, after a new layer is expanded, the model evaluates the detection performance of the current model on the verification data set, and compared with the last evaluation, if the current performance is improved to be smaller than a set threshold value, the model stops updating;
s204, after the detection is finished, a fraud detection result of the transaction data is obtained.
Preferably, the multi-presentation structure obtaining the characterization learning vector includes the steps of:
s401 and initialization structure
Setting initialization parameters of a multi-presentation structure: initializing the structure and number of Autoencoder models;
s402, setting BaggingBalance
Initializing parameters of the baggingbalane method: extracting the number of data sets and the number of features in the feature subsets;
s403, dividing the data set
Inputting transaction data, and dividing the transaction data by using a bagingengalance method;
s404, acquiring a sampling data set
Dividing the data set, sampling the data set, and obtaining a sampling data set required by an Autoencoder model;
s405, training Autoencoder
Acquiring a sampling data set as an input of an Autoencoder model, and training the Autoencoder model;
s406, generating a characterization learning vector
And training the Autoencoder model, and obtaining the vector of the middle hidden layer as the characterization learning vector of the final data set.
The invention relates to a transaction fraud behavior depth detection method based on characterization learning, which is characterized in that an Autoencoder-based characterization learning multi-presentation structure and a sample imbalance processing mechanism based on BaggingBase are provided. The electronic transaction fraud detection method provided by the invention can effectively detect fraud in electronic transactions. From the practical point of view, the method provided by the invention establishes a framework of an electronic transaction fraud detection method by representing and learning a multi-presentation structure and a sample imbalance processing mechanism based on a BaggingBalance, and provides technical support for solving fraud transaction detection.
Drawings
FIG. 1 is an overall framework of a fraud depth detection method for electronic transactions, the detection system consisting essentially of two structures, a characterization learning structure multi-presentation and an OSD-DF classification structure;
FIG. 2 is a specific flow chart of the present invention;
FIG. 3 is a flow chart of a sample imbalance processing mechanism based on BaggingBalance;
FIG. 4 is a flow chart of an Autoencoder-based token learning multi-presentation architecture;
fig. 5 is a flow chart of the simulation of the electronic transaction of the bank.
Detailed Description
The invention will be further illustrated with reference to specific examples. It is to be understood that these examples are illustrative of the present invention and are not intended to limit the scope of the present invention. Further, it is understood that various changes and modifications may be made by those skilled in the art after reading the teachings of the present invention, and such equivalents are intended to fall within the scope of the claims appended hereto.
The network transaction fraud detection method related by the invention mainly comprises the following three parts:
(1) An electronic transaction fraud detection method based on characterization learning. The detection method mainly comprises a characteristic learning structure multi-presentation structure and an OSD-DF classification structure.
(2) Sample imbalance processing mechanism based on bagingengance method. The Bagging balance method based on the Bagging idea is provided, and the problem of sample imbalance in transaction data is solved on data input.
(3) The multi-presentation structure is learned based on the representation of the Autoencoder model. And a multi-reproduction structure is provided, an Autoencoder model is introduced, and the characteristic learning capacity of the detection model is enhanced.
The technical core of the invention is that the part (2) and the part (3) are provided with a network transaction fraud detection method. In the task of fraud transaction detection, the key to the implementation of the electronic transaction fraud detection method is that: on one hand, the concealment of the fraudulent transaction can seriously influence the detection effect of the fraudulent detection model, and the characteristic learning capability of the detection method is very important for detecting the fraudulent transaction; meanwhile, sample imbalance of electronic transaction fraud detection also affects the model detection effect. Aiming at the concealment of fraudulent activity and sample imbalance in electronic transaction, the invention provides a sample imbalance processing mechanism based on a baggingbalane method and a representation learning multi-presentation structure based on an Autoencoder model, and establishes the electronic transaction fraud detection method.
According to the above ideas, the framework for establishing the electronic fraud transaction detection method of the present invention is as follows:
(1) The multi-presentation structure is learned based on the representation of the Autoencoder model. And inputting transaction data, extracting a data set from the transaction data by a bagmingbalance method, and inputting the data set into a corresponding Autoencoder model to obtain a characterization learning vector corresponding to the data set. The specific operation steps are as follows:
s101, extracting training set by using Baggingbalance method
And the Baggingbalance method is adopted to downsample the transaction data, balance the input data set and solve the problem of unbalanced samples.
S102, autoencoder model training
After the data set is obtained through the bagingbalance method, the data set is input into a corresponding Autoencoder model to carry out model training.
S103, obtaining characterization learning vectors
After the Autoencoder model is trained, the hidden layer vector in the model is extracted and used as the characterization learning vector of the training set.
(2) OSD-DF classification structure. The characteristic learning vector obtained by the multi-presentation structure is used as the input of the OSD-DF classifying structure, the OSD-DF classifying structure is trained, and the fraud detection result of the transaction data is obtained. The specific operation steps are as follows:
s201, data input
A token learning vector of the multi-presentation structure is obtained as an input to the structure.
S202, initializing a model
Setting a model structure, inputting data into a processing layer of a model, and outputting a processing result to a next layer;
s203, training model
The next layer receives the feature information processed by the previous layer, and repeats the operation of step S202 to continue expanding the next layer. When a new layer is extended, the model will evaluate the detection performance of the current model on the verification dataset. If the current performance improvement is less than the set threshold, the model stops updating as compared to the previous evaluation.
And S204, after the detection is finished, obtaining a fraud detection result of the transaction data.
In the framework of establishing an electronic transaction fraud detection method, the invention firstly provides a sample imbalance processing mechanism based on a BaggingBalance method, which is used for solving the problem of sample imbalance in transaction data. The specific operation steps of the mechanism are as follows:
s301, dividing data set
Dividing transaction data into sample sets D of legal transactions major And sample set D of fraudulent transactions minor 。
S302, data downsampling
Data set D for each random sampling legal transaction major Selecting a sample copy to be placed in the data set D sample The sample is then placed back into data set D major Is a kind of medium. Sample set D for sampling data volume as fraudulent transactions minor Is of a size of (a) and (b).
S303, feature sampling
And randomly selecting a feature subset from the feature space, and performing feature filtering on the downsampled data.
S304, returning the data set.
Aiming at feature sparsity caused by sparse samples in fraud detection, the invention provides a multi-representation structure based on representation learning of an Autoencoder model, further enhances the representation learning capacity of a detection method, and simultaneously combines the proposed bagging balance method to alleviate the problem caused by sample imbalance in electronic transaction. The specific flow of the multi-presentation structure is as follows:
s401 and initialization structure
Setting initialization parameters of a structure: initialization structure and number of Autoencoder model.
S402, setting Baggingbalance method
Initializing parameters of the baggingbalane method: the number of data sets and the number of features in the feature subset are extracted.
S403, dividing the data set
The original data set is input, and the original input data set is divided by using a bagingengalance method.
S404 acquiring a sample dataset
And after the data set is divided, sampling the data set to obtain a sampling data set required by an Autoencoder model.
S405, training Autoencoder model
And acquiring a sampling data set as an input of an Autoencoder model, and training the Autoencoder model.
S406 generates a token learning vector
And training the Autoencoder model, and obtaining the vector of the middle hidden layer as the characterization learning vector of the final data set.
The method and the system for detecting the network transaction fraud depth can be applied to large-scale network service systems such as a certain bank, and the method and the system for detecting the network transaction fraud depth are applied to a certain domestic mainstream bank transaction system for experimental verification, and realize a bank transaction data real-time fraud detection flow through a system reconstruction method. The specific method describes fig. 5.
Taking the example of real electronic transaction data of a mainstream bank in China, the data set comprises B2C transaction records of three months of banks, wherein about 7 ten thousand transactions are marked as fraudulent transactions:
s1, selecting transaction data of the first two months as a training set, and the last month as simulation test real-time transaction data.
S2, setting the number of the sampling data sets, and extracting the sampling data sets from the original data sets by using a bagging balance method.
S3, setting the number of Autoencoder models, training the Autoencoder models by using a sampling data set, and obtaining a characterization vector of a multi-presentation structure as input of an OSD-DF classification structure.
S4, training an OSD-DF classification structure, and performing fraud detection and marking on the transaction data.
S5, inputting the simulated test real-time transaction data into a trained transaction detection method, carrying out transaction detection according to a simulated transaction flow shown in FIG. 5, and judging whether the transaction data are fraudulent.
Claims (2)
1. The transaction fraud behavior depth detection method based on characterization learning is characterized by comprising the following steps of:
(1) Establishing a representation learning multi-presentation structure based on an Autoencoder model, extracting a data set from transaction data through a bagging balance method, inputting the data set into the multi-presentation structure to obtain a corresponding representation learning vector, wherein the extracting the data set from the transaction data through the bagging balance method comprises the following steps of:
s301, dividing data set
Dividing transaction data into sample sets D of legal transactions malor And sample set D of fraudulent transactions minor ；
S302, data downsampling
Sample set D is randomly sampled every time major Selecting a sample copy to be placed in the data set D sample Then put the sample back into sample set D major In which the sampled data volume is the sample set D minor Is of a size of (2);
s303, feature sampling
Randomly selecting a feature subset from the feature space, and performing feature filtering on the downsampled data;
s304, returning a data set;
(2) An OSD-DF classifying structure is established, a multi-presenting structure is used for obtaining a characterization learning vector as the input of the OSD-DF classifying structure, the OSD-DF classifying structure is trained, and a fraud detection result of transaction data is obtained, comprising the following steps:
s201, data input
Acquiring a characterization learning vector obtained by a multi-presentation structure as an input of an OSD-DF classification structure;
s202, initializing a model
Setting a model structure of an OSD-DF classifying structure, inputting data into a processing layer of the OSD-DF classifying structure, and outputting a processing result to a next layer;
s203, training model
The next layer receives the characteristic information processed by the previous layer, the operation of the step S202 is repeated, the next layer is continuously expanded, after a new layer is expanded, the model evaluates the detection performance of the current model on the verification data set, and compared with the last evaluation, if the current performance is improved to be smaller than a set threshold value, the model stops updating;
s204, after the detection is finished, a fraud detection result of the transaction data is obtained.
2. The transaction fraud depth detection method based on token learning according to claim 1, wherein the multi-presentation structure obtaining token learning vector comprises the steps of:
s401 and initialization structure
Setting initialization parameters of a multi-presentation structure: initializing the structure and number of Autoencoder models;
s402, setting BaggingBalance
Initializing parameters of the baggingbalane method: extracting the number of data sets and the number of features in the feature subsets;
s403, dividing the data set
Inputting transaction data, and dividing the transaction data by using a bagingengalance method;
s404, acquiring a sampling data set
Dividing the data set, sampling the data set, and obtaining a sampling data set required by an Autoencoder model;
s405, training Autoencoder
Acquiring a sampling data set as an input of an Autoencoder model, and training the Autoencoder model;
s406, generating a characterization learning vector
And training the Autoencoder model, and obtaining the vector of the middle hidden layer as the characterization learning vector of the final data set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910327470.1A CN110084609B (en) | 2019-04-23 | 2019-04-23 | Transaction fraud behavior deep detection method based on characterization learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910327470.1A CN110084609B (en) | 2019-04-23 | 2019-04-23 | Transaction fraud behavior deep detection method based on characterization learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110084609A CN110084609A (en) | 2019-08-02 |
| CN110084609B true CN110084609B (en) | 2023-06-02 |
Family
ID=67416228
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910327470.1A Active CN110084609B (en) | 2019-04-23 | 2019-04-23 | Transaction fraud behavior deep detection method based on characterization learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110084609B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111275098A (en) * | 2020-01-17 | 2020-06-12 | 同济大学 | Encoder-LSTM deep learning model applied to credit card fraud detection and method thereof |
| CN111401908A (en) * | 2020-03-11 | 2020-07-10 | 支付宝(杭州)信息技术有限公司 | Transaction behavior type determination method, device and equipment |
| CN112884030B (en) * | 2021-02-04 | 2022-05-06 | 重庆邮电大学 | Cross reconstruction based multi-view classification system and method |
| CN116562901B (en) * | 2023-06-25 | 2024-04-02 | 福建润楼数字科技有限公司 | Automatic generation method of anti-fraud rule based on machine learning |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8726379B1 (en) * | 2011-07-15 | 2014-05-13 | Norse Corporation | Systems and methods for dynamic protection from electronic attacks |
| CN105760889A (en) * | 2016-03-01 | 2016-07-13 | 中国科学技术大学 | Efficient imbalanced data set classification method |
| CN107590690A (en) * | 2017-09-05 | 2018-01-16 | 阿里巴巴集团控股有限公司 | Data processing method, device and server |
| CN108304884A (en) * | 2018-02-23 | 2018-07-20 | 华东理工大学 | A kind of cost-sensitive stacking integrated study frame of feature based inverse mapping |
| CN109376179A (en) * | 2018-08-24 | 2019-02-22 | 苏宁消费金融有限公司 | A kind of sample equilibrating method in data mining |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11205103B2 (en) * | 2016-12-09 | 2021-12-21 | The Research Foundation for the State University | Semisupervised autoencoder for sentiment analysis |
-
2019
- 2019-04-23 CN CN201910327470.1A patent/CN110084609B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8726379B1 (en) * | 2011-07-15 | 2014-05-13 | Norse Corporation | Systems and methods for dynamic protection from electronic attacks |
| CN105760889A (en) * | 2016-03-01 | 2016-07-13 | 中国科学技术大学 | Efficient imbalanced data set classification method |
| CN107590690A (en) * | 2017-09-05 | 2018-01-16 | 阿里巴巴集团控股有限公司 | Data processing method, device and server |
| CN108304884A (en) * | 2018-02-23 | 2018-07-20 | 华东理工大学 | A kind of cost-sensitive stacking integrated study frame of feature based inverse mapping |
| CN109376179A (en) * | 2018-08-24 | 2019-02-22 | 苏宁消费金融有限公司 | A kind of sample equilibrating method in data mining |
Non-Patent Citations (3)
| Title |
|---|
| Towards the Classification of Cancer Subtypes by Using Cascade Deep Forest Model in Gene Expression Data;Yang Guo,et al;《2017 IEEE International Conference on Bioinformatics and Biomedicine (BIBM)》;20171218;正文第1664-1669页 * |
| 基于集成学习的二手房产数据分类研究;宋健;《中国优秀硕士学位论文全文数据库》;20181015;J145-201 * |
| 基于集成特征选择的点击欺诈检测方法;郭汉 等;《计算机工程与应用》;20111003;正文第246-251页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110084609A (en) | 2019-08-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110084609B (en) | Transaction fraud behavior deep detection method based on characterization learning | |
| TWI789345B (en) | Modeling method and device for machine learning model | |
| WO2018103456A1 (en) | Method and apparatus for grouping communities on the basis of feature matching network, and electronic device | |
| CN109034194B (en) | Transaction fraud behavior deep detection method based on feature differentiation | |
| WO2019179403A1 (en) | Fraud transaction detection method based on sequence width depth learning | |
| CN108734380B (en) | Risk account determination method and device and computing equipment | |
| CN110213244A (en) | A kind of network inbreak detection method based on space-time characteristic fusion | |
| WO2022121145A1 (en) | Ethereum phishing scam detection method and apparatus based on graph classification | |
| CN111539733B (en) | Fraud transaction identification method, system and device based on full-center loss function | |
| WO2021037280A2 (en) | Rnn-based anti-money laundering model training method, apparatus and device, and medium | |
| CN110084610A (en) | A kind of network trading fraud detection system based on twin neural network | |
| CN111325619A (en) | Credit card fraud detection model updating method and device based on joint learning | |
| Li et al. | One-class knowledge distillation for face presentation attack detection | |
| Jonnalagadda et al. | Credit card fraud detection using Random Forest Algorithm | |
| CN113052577A (en) | Method and system for estimating category of virtual address of block chain digital currency | |
| Choudhury et al. | An efficient way to detect credit card fraud using machine learning methodologies | |
| CN116993513A (en) | Financial wind control model interpretation method and device and computer equipment | |
| CN112365352A (en) | Anti-cash-out method and device based on graph neural network | |
| CN114818999B (en) | Account identification method and system based on self-encoder and generation countermeasure network | |
| CN116821688A (en) | Method for processing data set in credit card fraud transaction based on clustering downsampling technology | |
| CN116485406A (en) | Account detection method and device, storage medium and electronic equipment | |
| CN114240659A (en) | Block chain abnormal node identification method based on dynamic graph convolutional neural network | |
| Chauhan et al. | Bertops: Studying bert representations under a topological lens | |
| Huang et al. | Improved Deep Forest Mode for Detection of Fraudulent Online Transaction. | |
| Jose et al. | Detection of Credit Card Fraud Using Resampling and Boosting Technique |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |