CN106682067B - A kind of anti-fake monitoring system of machine learning based on transaction data - Google Patents
A kind of anti-fake monitoring system of machine learning based on transaction data Download PDFInfo
- Publication number
- CN106682067B CN106682067B CN201610981804.3A CN201610981804A CN106682067B CN 106682067 B CN106682067 B CN 106682067B CN 201610981804 A CN201610981804 A CN 201610981804A CN 106682067 B CN106682067 B CN 106682067B
- Authority
- CN
- China
- Prior art keywords
- data
- engine
- model
- sample
- algorithm
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/21—Design, administration or maintenance of databases
- G06F16/215—Improving data quality; Data cleansing, e.g. de-duplication, removing invalid entries or correcting typographical errors
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/25—Integrating or interfacing systems involving database management systems
- G06F16/254—Extract, transform and load [ETL] procedures, e.g. ETL data flows in data warehouses
-
- 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
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/04—Trading; Exchange, e.g. stocks, commodities, derivatives or currency exchange
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Databases & Information Systems (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Development Economics (AREA)
- General Business, Economics & Management (AREA)
- Technology Law (AREA)
- Strategic Management (AREA)
- Marketing (AREA)
- Economics (AREA)
- Quality & Reliability (AREA)
- Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of anti-fake monitoring system of machine learning based on transaction data, including management platform, ETL module, sample engine, stream process engine, training engine, prediction engine and decision engine；Stream process engine handles rapid extraction and the calculating that feature is carried out to huge transaction initial data by streaming big data, and representational feature is obtained from magnanimity initial data, fully extracts the information in data.Model training module uses a variety of machine learning models optimized for monetary losses rate, black sample recall ratio and integrated learning framework, what is obtained is the composite model for some index optimization, overcome over-fitting that single model brings, it is unstable the defects of, improve the stability and generalization ability of model；Model training module is by pre-set renewal time, again automatic acquisition latest data and training pattern, so that model remains validity, the model Problem of Failure for avoiding fraud variation from bringing.
Description
Technical field
The present invention relates to financial field, refers in particular to a kind of anti-fake monitoring system of machine learning based on transaction data.
Background technology
The booming financial revolution for creating a new round of Internet technology, but too fast growth is also contained greatly
Blindness, what is be accompanied is the risk of fraud being on the rise.Fake monitoring pattern relatively conventional at present is included based on big number
According to risk policy, anti-fraud system and elite air control team etc..Although most of payment mechanisms have fake monitoring system,
But majority still relies on elite team and rule induction is carried out on the basis of analysis of cases.However, fraudulent mean emerge in an endless stream and
The inconsistent of trading activity brings difficulty to rule induction.Meanwhile current algorithm is difficult to keep its robustness, performance
It will decline with the expansion of rule system, can not ensure there is high recall ratio while high precision ratio, so as to reduce user experience.
Machine learning is less dependent on manual analysis again since it is in non-linear and cost-sensitive scene advantage,
More preferably robustness and stability are shown, so being increasingly becoming a kind of new fraud detection scheme.
The content of the invention
It is an object of the invention in view of the problems of the existing technology, there is provided a set of transaction swindling towards financial field
Real-time monitoring system.By carrying out analysis and modeling to the historical trading data after cleaning, when New Transaction occurs, will currently hand over
Easy is compared with historical trading behavior, real-time judge is carried out to the transaction risk according to the scoring of output, so as to reach
To the target of real-time deal fraud detection.The system can reach higher precision and look into complete in the case of relatively low rate of false alarm
Rate, so as to ensure the transaction security of client.
The purpose of the present invention is what is be achieved through the following technical solutions：A kind of machine learning based on transaction data is counter to take advantage of
Cheat monitoring system, the system include management platform, ETL module, sample engine, stream process engine, training engine, prediction engine and
Decision engine；
The management platform provides the configuration information of each module, concurrent pattern drawing train request and predictions request, to mould
Type is managed and updates operation；The configuration information includes the data time section needed for ETL module, needed for sample engine
Database field, feature name and calculation needed for stream process engine, algorithm title and algorithm ginseng needed for training engine
Number.
The ETL module extracts raw data base data, carries out data pick-up, turns according to the configuration information of management platform
Change, in-stockroom operation；The data transformation operations are mainly cleaned and standardized to data, including two parts：By initial data
The customized data in storehouse are converted into normal data；The field that machine learning model can not be handled is converted；Data loading is grasped
Make the data being disposed being stored in any frequently-used data storehouse.
The sample engine samples initial data according to the configuration information of management platform, is extracted from initial data
The Database field that stream process engine needs.
The feature name and calculation that the stream process engine is configured according to management platform, feature is carried out to sampled data
Extraction and calculating.
The trained engine includes data cleansing, model training, model evaluation；The data cleansing, lacks data
The normal data cleaning operations such as the processing of mistake value, normalized；The model training, according to the model parameter of setting, using clear
Characteristic after washing is trained, and is specially：The algorithm title and algorithm parameter of management platform configuration are read, is called common
Machine learning algorithm, includes supervision algorithm and unsupervised algorithm is learnt；There is supervision algorithm to include logistic regression, linearly return
Return, support vector machines, decision Tree algorithms etc.；Unsupervised algorithm includes k-means clusters etc.；The model evaluation, using new
Data set evaluates trained model, according to indexs such as the recall ratio of output, precision ratio, KS values, ROC curves to model
Quality is evaluated, if satisfactory quality can carry out model deployment and use；The model training module passes through advance
It is the renewal time of setting, automatic to obtain latest data and again training pattern, so that model remains validity.
The prediction engine calls trained model to flowing successively through the reality of ETL module, sample engine, stream process engine
Border transaction data is differentiated that output belongs to the probability of arm's length dealing and belongs to the probability of fraudulent trading, and prediction result is transmitted to
Decision engine.
The decision engine carries out decision-making according to the output of prediction engine to the dangerous of the transaction.
Further, the machine learning algorithm in the trained engine, is transformed, specifically for black sample recall ratio
For：The weight bigger than white sample is assigned to black sample losses function, makes it be more likely to find out more black samples；It is alternatively, right
Black sample carries out over-sampling, and white sample carries out lack sampling；Alternatively, increasing regular terms after loss function, model complexity is reduced,
Improve model generalized ability；Alternatively, using integrated study frame, the over-fitting of single model is overcome.
Transformed for monetary losses rate, the big weight of the low amount of money is assigned to high amount of money sample, is more likely to model
The high amount of money sample of few misclassification；Alternatively, according to single transaction amount dynamic adjustment probability threshold value, make the transaction to the high amount of money be more difficult to by
It is determined as white sample；
Do and optimize for algorithm performance, using the function that can be performed in GPU accelerating algorithms with parallelization, substantially reduce training
And predicted time；Alternatively, the calculating for realizing algorithm bottom using linear algebra storehouse operates；It is alternatively, parallel using multithreading
Algorithm is realized in change.
Further, stream process engine is handled by streaming big data and carries out the fast of feature to huge transaction initial data
Speed extraction and calculate, can obtain in some time interval certain user's history trading volume under some dimension add up, accounting, variance,
The characteristic quantities such as average, summation, counting, minimum number statistics, standard deviation statistics calculating, the degree of bias, kurtosis, duplicate removal.
Two parts of training and prediction can be divided into whole system process for using.
During training, the information of modules is configured using management platform, and initiates train request, ETL module root
According to configuration information, raw data base data are extracted, carry out data pick-up, conversion, in-stockroom operation.Sample engine is according to configuration to original
Beginning data are sampled, the Database field needed.Stream process engine carries out sampled data feature extraction and calculating, instruction
Practice engine first to clean data, according to the model parameter of setting, be trained using characteristic, then utilized newly
Data the set pair analysis model is assessed, according to multiple indexes judgment models quality, if satisfactory quality can carry out mold portion
Administration and use, so far training part is terminated, otherwise repeatedly aforesaid operations process.
During prediction, ETL module obtains transaction data, sample engine and Liu Chu in real time according to the configuration of gathered data during training
Engine is managed by sampling operation and streaming computing, obtains characteristic and input model, prediction engine obtains model output, decision-making
Engine carries out Real-time Decision according to output probability.
The system contrasts the prior art and system has obvious advantage, and system can maintain better stability/robustness
While, ensure higher recall ratio and relatively low rate of false alarm.Above-mentioned characteristic is mainly by following several promises：Stream process engine leads to
The processing of overflow-type big data carries out huge transaction initial data rapid extraction and the calculating of feature, from magnanimity initial data
Representational feature is obtained, fully extracts the information in data.Model training module is using a variety of for monetary losses rate, black
The machine learning model and integrated learning framework that sample recall ratio optimized, what is obtained is the compound die for some index optimization
Type, overcome over-fitting that single model brings, it is unstable the defects of, improve the stability and generalization ability of model；Model
Training module passes through pre-set renewal time, automatic acquisition latest data and again training pattern, so that model is all the time
Validity is kept, the model Problem of Failure for avoiding fraud variation from bringing.
Brief description of the drawings
Fig. 1 is the structure diagram of the preferred embodiments of the invention.
Fig. 2 is exemplary timing diagram in the preferred embodiments of the invention.
Embodiment
More clearly to illustrate the architectural feature and effect of the present invention, come below in conjunction with the accompanying drawings with specific embodiment to this hair
It is bright to be described in detail.
As shown in Figure 1, 2, the anti-fake monitoring system of a kind of machine learning based on transaction data provided by the invention, including
Management platform, ETL module, sample engine, stream process engine, training engine, prediction engine and decision engine；
Management platform provides the visualization interface of system administration, and user can be flat in management by the information that each module needs
Configured on platform, each module will obtain configuration information from management platform automatically and carry out respective operations.Management platform may be used also
To initiate model training request and predictions request, model is managed and updates operation.
After train request is received, ETL module obtains the trading activity data of financial system front end triggering, carries out data pumping
Take, change, in-stockroom operation.Specifically, which mainly obtains the data of financial system trading activity, including during transaction
Between, loco, transaction IP, terminal type (movement, PC ends, operating system classification etc.), transaction amount, Transaction Account number etc., this
A little data can be divided mainly into following major class：
1st, trading environment：Including exchange hour, transaction IP, transaction terminal etc..
2nd, transaction content：Including transaction amount, transaction account number, trading password etc..
3rd, account number feature：Including regional feature, space-time characteristic, sex character, age characteristics etc..
4th, aggregated data：Refer to the polymerization amount of data, including 3 it is small when interior transaction count etc..
5th, other data：Refer to the data with the other side of the account relating.
Data transformation operations are mainly cleaned and standardized to data, mainly including two parts：By raw data base certainly
The data of definition are converted into normal data, such as the time will be converted into the standard time；The field that machine learning model can not be handled
Converted, such as such as telephone number is converted to ownership place.
Data loading operation is exactly that the data being disposed are stored in any frequently-used data storehouse, such as Oracle.
Sample engine takes the data of needs by the configuration file of management platform from above-mentioned database, and configuration file includes
The period of required data, the information such as title of required field, equivalent to list needed for portion, the data got are stored in
In memory.
Stream process engine calculates the data that sample engine is got, according to the characteristic information needed in management platform,
Initial data is converted into characteristic by engine, as certain is characterized in calculating accumulative trade gold of each user when the past 24 is small
Volume, stream process engine will search each user past 24 it is small when transaction record and transaction amount is added up.Final meter
Hereof, file can be arbitrary standards form, such as CSV, txt for good result storage.
Training engine includes data cleansing, model training, model evaluation.Missing values processing, normalizing are carried out to data first
The normal data cleaning operations such as change processing.Then the algorithm title configured on management platform interface and algorithm parameter are read, is called
Common machine learning algorithm, includes supervision algorithm and unsupervised algorithm is learnt.Have supervision algorithm include logistic regression,
Linear regression, support vector machines, decision Tree algorithms etc.；Unsupervised algorithm includes k-means clusters etc..
These algorithms are transformed for black sample recall ratio, are specially：It is assigned to black sample losses function than white sample
Big weight, makes it be more likely to find out more black samples；Alternatively, carrying out over-sampling to black sample, white sample carries out owing to adopt
Sample；Alternatively, increasing regular terms after loss function, model complexity is reduced, improves model generalized ability；Alternatively, using integrated
Learning framework, overcomes the over-fitting of single model.
Transformed for monetary losses rate, the big weight of the low amount of money is assigned to high amount of money sample, is more likely to model
The high amount of money sample of few misclassification；Alternatively, according to single transaction amount dynamic adjustment probability threshold value, make the transaction to the high amount of money be more difficult to by
It is determined as white sample；
Do and optimize for algorithm performance, using the function that can be performed in GPU accelerating algorithms with parallelization, substantially reduce training
And predicted time；Alternatively, the calculating for realizing algorithm bottom using linear algebra storehouse operates；It is alternatively, parallel using multithreading
Algorithm is realized in change.
Parameter is adjusted, obtains the model for meeting the index requests such as accuracy rate, recall rate, and use test the set pair analysis model
Assessed, whether observing and nursing can extensive to other data set.Information in training process feeds back to management platform.Finally
The model write-in file that training finishes carries out persistence.Production environment deployment will be carried out for suitable model, makes truly to hand over
Easy data flow through whole system and carry out real-time blocking to possible risk.At the same time, model training module can also pass through
It is pre-set renewal time, automatic to obtain latest data and the suitable model of re -training, so that model has remained
Effect property.
Prediction engine and decision engine play a role after model actual deployment, real trade data in units of bar successively
ETL module, sample engine are flowed through, stream process engine simultaneously carries out the friendship after same operation, handled well in above-mentioned training process
Easy data directly input prediction engine, and prediction engine calls trained model to differentiate this data, and output belongs to just
The probability often merchandised and the probability for belonging to fraudulent trading, decision engine is transmitted to by prediction result.Decision engine is according to prediction engine
Output, to when transaction carry out Real-time Decision.
The design focal point of the present invention is：Overall gui interface and managing configuration information are provided by management platform, is passed through
ETL module carries out data quickly to change, is put in storage, and extensive raw data set is obtained using sampling module, big by streaming
Data processing carries out huge transaction initial data rapid extraction and the calculating of feature, and generation has been obtained from magnanimity initial data
The feature of table.Machine learning algorithm optimizes by a variety of for monetary losses rate, black sample recall ratio, reasonable by setting
Algorithm parameter, train outstanding model, and the assessment of multiple data sets is carried out to this model.Designed more than, this is
System can carry out accurate decision-making to transaction in real time.
The above described is only a preferred embodiment of the present invention, be not intended to limit the scope of the present invention,
Therefore any subtle modifications, equivalent variations and modifications that every technical spirit according to the present invention makees above example, still
Belong in the range of technical solution of the present invention.
Claims (2)
- A kind of 1. anti-fake monitoring system of machine learning based on transaction data, it is characterised in that the system include management platform, ETL module, sample engine, stream process engine, training engine, prediction engine and decision engine；The management platform provides the configuration information of each module, concurrent pattern drawing train request and predictions request, to model into Row management and renewal operation；The configuration information includes the data time section needed for ETL module, the data needed for sample engine Storehouse field, feature name and calculation needed for stream process engine, algorithm title and algorithm parameter needed for training engine；The ETL module extracts raw data base data, carries out data pick-up, change, enter according to the configuration information of management platform Storehouse operates；The data transformation operations are mainly cleaned and standardized to data, including two parts：Raw data base is made by oneself The data of justice are converted into normal data；The field that machine learning model can not be handled is converted；Data loading operation will place Manage the data finished and be stored in any frequently-used data storehouse；The sample engine samples initial data according to the configuration information of management platform, is extracted from initial data at stream Manage the Database field that engine needs；The feature name and calculation that the stream process engine is configured according to management platform, feature extraction is carried out to sampled data And calculating；The trained engine includes data cleansing, model training, model evaluation；Data are carried out missing values by the data cleansing Processing, normalized；The model training, according to the model parameter of setting, is trained using the characteristic after cleaning, Specially：The algorithm title and algorithm parameter of management platform configuration are read, common machine learning algorithm is called, includes supervision Algorithm and unsupervised algorithm are learnt；There is supervision algorithm to include logistic regression, linear regression, support vector machines, decision tree calculation Method；Unsupervised algorithm is clustered including k-means；The model evaluation, comments trained model using new data set Valency, according to the recall ratio of output, precision ratio, KS values, ROC curve index evaluates model quality, if quality conforms to Model deployment and use can be carried out by asking；The model training module is obtained newest automatically by pre-set renewal time Data and again training pattern, so that model remains validity；Machine learning algorithm in the trained engine, is transformed for black sample recall ratio, is specially：To black sample losses Function is assigned to the weight bigger than white sample, makes it be more likely to find out more black samples；Alternatively, black sample was adopted Sample, white sample carry out lack sampling；Alternatively, increasing regular terms after loss function, model complexity is reduced, improves model generalized energy Power；Alternatively, using integrated study frame, the over-fitting of single model is overcome；Transformed for monetary losses rate, the weight bigger than the low amount of money is assigned to high amount of money sample, be more likely to model few The high amount of money sample of misclassification；Alternatively, according to single transaction amount dynamic adjustment probability threshold value, the transaction to the high amount of money is set to be more difficult to be judged to Wei not white sample；Do and optimize for algorithm performance, using the function for being capable of parallelization execution in GPU accelerating algorithms, substantially reduce trained and pre- Survey the time；Alternatively, the calculating for realizing algorithm bottom using linear algebra storehouse operates；It is alternatively, real using multithreading parallelization Existing algorithm；The prediction engine calls trained model to flowing successively through the actual friendship of ETL module, sample engine, stream process engine Easy data are differentiated that output belongs to the probability of arm's length dealing and belongs to the probability of fraudulent trading, and prediction result is transmitted to decision-making Engine；The decision engine carries out decision-making according to the output of prediction engine to the dangerous of the transaction.
- A kind of 2. anti-fake monitoring system of machine learning based on transaction data according to claim 1, it is characterised in that Stream process engine handles rapid extraction and the calculating that feature is carried out to huge transaction initial data by streaming big data, obtains In some time interval under some dimension certain user's history trading volume add up, accounting, variance, average, summation, counting, minimum number Statistics, standard deviation statistics calculating, the degree of bias, kurtosis, duplicate removal characteristic quantity.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610981804.3A CN106682067B (en) | 2016-11-08 | 2016-11-08 | A kind of anti-fake monitoring system of machine learning based on transaction data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610981804.3A CN106682067B (en) | 2016-11-08 | 2016-11-08 | A kind of anti-fake monitoring system of machine learning based on transaction data |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106682067A CN106682067A (en) | 2017-05-17 |
| CN106682067B true CN106682067B (en) | 2018-05-01 |
Family
ID=58839430
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610981804.3A Active CN106682067B (en) | 2016-11-08 | 2016-11-08 | A kind of anti-fake monitoring system of machine learning based on transaction data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106682067B (en) |
Families Citing this family (36)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180350006A1 (en) * | 2017-06-02 | 2018-12-06 | Visa International Service Association | System, Method, and Apparatus for Self-Adaptive Scoring to Detect Misuse or Abuse of Commercial Cards |
| CN107785058A (en) * | 2017-07-24 | 2018-03-09 | 平安科技（深圳）有限公司 | Anti- fraud recognition methods, storage medium and the server for carrying safety brain |
| CN107563884A (en) * | 2017-07-28 | 2018-01-09 | 浙江邦盛科技有限公司 | A kind of method for reducing loss of assets rate |
| CN107566358B (en) * | 2017-08-25 | 2020-10-30 | 腾讯科技（深圳）有限公司 | Risk early warning prompting method, device, medium and equipment |
| CN108053318B (en) * | 2017-12-20 | 2021-09-03 | 奇安信科技集团股份有限公司 | Method and device for identifying abnormal transactions |
| CN108122114A (en) * | 2017-12-25 | 2018-06-05 | 同济大学 | For abnormal repeat business fraud detection method, system, medium and equipment |
| CN108230140A (en) * | 2017-12-29 | 2018-06-29 | 阿里巴巴集团控股有限公司 | The method and apparatus of pushed information, the method and apparatus for determining input default value |
| CN108446978A (en) * | 2018-02-12 | 2018-08-24 | 阿里巴巴集团控股有限公司 | Handle the method and device of transaction data |
| CN108446720A (en) * | 2018-02-27 | 2018-08-24 | 华青融天（北京）技术股份有限公司 | Abnormal deviation data examination method and system |
| CN108038701A (en) * | 2018-03-20 | 2018-05-15 | 杭州恩牛网络技术有限公司 | A kind of integrated study is counter to cheat test method and system |
| CN110334720A (en) * | 2018-03-30 | 2019-10-15 | 百度在线网络技术（北京）有限公司 | Feature extracting method, device, server and the storage medium of business datum |
| CN108765135A (en) * | 2018-05-29 | 2018-11-06 | 佛山市金晶微阅信息科技有限公司 | A kind of credit is counter to cheat identifying system |
| CN108763573A (en) * | 2018-06-06 | 2018-11-06 | 众安信息技术服务有限公司 | A kind of OLAP engines method for routing and system based on machine learning |
| CN109086975B (en) * | 2018-07-10 | 2022-04-26 | 创新先进技术有限公司 | Transaction risk identification method and device |
| CN109308615B (en) * | 2018-08-02 | 2020-12-29 | 同济大学 | Real-time fraud transaction detection method, system, storage medium and electronic terminal based on statistical sequence characteristics |
| CN109344394B (en) * | 2018-08-30 | 2023-06-06 | 广发证券股份有限公司 | Text data quality monitoring system |
| CN109443419B (en) * | 2018-08-31 | 2021-08-31 | 广州致新电力科技有限公司 | Machine learning-based rectifier online monitoring method |
| CN109299434B (en) * | 2018-09-04 | 2019-06-14 | 重庆公共运输职业学院 | Cargo customs clearance big data is intelligently graded and sampling observation rate computing system |
| CN109447276B (en) * | 2018-09-17 | 2021-11-02 | 烽火通信科技股份有限公司 | Machine learning system, equipment and application method |
| CN109583729B (en) * | 2018-11-19 | 2023-06-20 | 创新先进技术有限公司 | Data processing method and device for platform online model |
| CN110009174B (en) * | 2018-12-13 | 2020-11-06 | 创新先进技术有限公司 | Risk recognition model training method and device and server |
| CN109635948A (en) * | 2018-12-19 | 2019-04-16 | 北京达佳互联信息技术有限公司 | On-line training method, apparatus, system and computer readable storage medium |
| CN109726239A (en) * | 2018-12-25 | 2019-05-07 | 厦门市美亚柏科信息股份有限公司 | The method, apparatus and readable storage medium storing program for executing that a kind of pair of forensic data is analyzed |
| CN110008984B (en) * | 2019-01-22 | 2023-07-25 | 创新先进技术有限公司 | Target fraud transaction model training method and device based on multitasking samples |
| TWI764767B (en) * | 2019-02-13 | 2022-05-11 | 華南商業銀行股份有限公司 | Automatic fund depositing system for planning customized deposit transfer amount |
| CN110111106A (en) * | 2019-05-06 | 2019-08-09 | 无线生活(北京)信息技术有限公司 | Transaction risk monitoring method and device |
| CN110888861A (en) * | 2019-11-12 | 2020-03-17 | 上海麦克风文化传媒有限公司 | Novel big data storage method |
| CN110956278A (en) * | 2019-11-26 | 2020-04-03 | 支付宝(杭州)信息技术有限公司 | Method and system for retraining machine learning models |
| CN110830515A (en) * | 2019-12-13 | 2020-02-21 | 支付宝(杭州)信息技术有限公司 | Flow detection method and device and electronic equipment |
| TWI760254B (en) * | 2020-02-13 | 2022-04-01 | 華南商業銀行股份有限公司 | Automatic fund depositing method using neural network |
| CN111461216B (en) * | 2020-03-31 | 2023-08-25 | 浙江邦盛科技股份有限公司 | Case risk identification method based on machine learning |
| CN111861699B (en) * | 2020-07-02 | 2021-06-22 | 北京睿知图远科技有限公司 | Anti-fraud index generation method based on operator data |
| CN112651785B (en) * | 2020-12-31 | 2023-12-08 | 中国农业银行股份有限公司 | Transaction amount real-time monitoring method and system |
| CN113506109A (en) * | 2021-07-27 | 2021-10-15 | 中国工商银行股份有限公司 | Fraud transaction identification method and device |
| CN115082076A (en) * | 2022-07-04 | 2022-09-20 | 北京天德科技有限公司 | Three-stage financial violation multiple judgment method based on block chain |
| CN116596532A (en) * | 2022-11-07 | 2023-08-15 | 北京天德科技有限公司 | Supervision method based on real-time suspicious transaction identification and supervision blockchain wallet |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103793484A (en) * | 2014-01-17 | 2014-05-14 | 五八同城信息技术有限公司 | Fraudulent conduct identification system based on machine learning in classified information website |
| CN104679777A (en) * | 2013-12-02 | 2015-06-03 | 中国银联股份有限公司 | Method and system for detecting fraudulent trading |
| CN106022508A (en) * | 2016-05-06 | 2016-10-12 | 陈丛威 | Method and apparatus for predicting user friend invitation behaviors of online financing platform |
-
2016
- 2016-11-08 CN CN201610981804.3A patent/CN106682067B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104679777A (en) * | 2013-12-02 | 2015-06-03 | 中国银联股份有限公司 | Method and system for detecting fraudulent trading |
| CN103793484A (en) * | 2014-01-17 | 2014-05-14 | 五八同城信息技术有限公司 | Fraudulent conduct identification system based on machine learning in classified information website |
| CN106022508A (en) * | 2016-05-06 | 2016-10-12 | 陈丛威 | Method and apparatus for predicting user friend invitation behaviors of online financing platform |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106682067A (en) | 2017-05-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106682067B (en) | A kind of anti-fake monitoring system of machine learning based on transaction data | |
| Liu et al. | Federaser: Enabling efficient client-level data removal from federated learning models | |
| CN108320043B (en) | Power distribution network equipment state diagnosis and prediction method based on electric power big data | |
| CN109816509A (en) | Generation method, terminal device and the medium of scorecard model | |
| WO2020077682A1 (en) | Service quality evaluation model training method and device | |
| CN107766929B (en) | Model analysis method and device | |
| CN108154430A (en) | A kind of credit scoring construction method based on machine learning and big data technology | |
| CN107742127A (en) | A kind of improved anti-electricity-theft intelligent early-warning system and method | |
| CN108898476A (en) | A kind of loan customer credit-graded approach and device | |
| TW201732662A (en) | Method and apparatus for establishing data identification model | |
| CN117240887B (en) | Wisdom thing networking energy management platform system | |
| CN107341688A (en) | The acquisition method and system of a kind of customer experience | |
| CN110119948A (en) | Based on when variable weight dynamic combined power consumer credit assessment method and system | |
| CN107862423A (en) | System evaluation method, intelligent evaluation system and computer-readable recording medium | |
| CN114048436A (en) | Construction method and construction device for forecasting enterprise financial data model | |
| CN110147389A (en) | Account number treating method and apparatus, storage medium and electronic device | |
| CN111798310A (en) | Rejection inference method based on Cox regression and logistic regression and electronic equipment | |
| CN109101296A (en) | Distributed auto deduction method and system | |
| CN109272312A (en) | Method and apparatus for transaction risk detecting real-time | |
| CN114140248A (en) | AI artificial intelligence technology-based abnormal transaction identification method | |
| CN114004654A (en) | Internet card user portrait construction method and system based on big data analysis | |
| CN111666978A (en) | Intelligent fault early warning system for IT system operation and maintenance big data | |
| CN110276299A (en) | Batch meter appearance fault picture identification model and method | |
| Luo et al. | Stock market forecasting algorithm based on improved neural network | |
| CN110059752A (en) | A kind of statistical learning querying method based on comentropy Sampling Estimation |
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 | ||
| CP01 | Change in the name or title of a patent holder | ||
| CP01 | Change in the name or title of a patent holder |
Address after: Room ABCD, 17th floor, building D, Paradise Software Park, No.3 xidoumen Road, Xihu District, Hangzhou City, Zhejiang Province, 310012 Patentee after: Zhejiang Bangsheng Technology Co.,Ltd. Address before: Room ABCD, 17th floor, building D, Paradise Software Park, No.3 xidoumen Road, Xihu District, Hangzhou City, Zhejiang Province, 310012 Patentee before: ZHEJIANG BANGSUN TECHNOLOGY Co.,Ltd. |