CN105095238B - For detecting the decision tree generation method of fraudulent trading - Google Patents
For detecting the decision tree generation method of fraudulent trading Download PDFInfo
- Publication number
- CN105095238B CN105095238B CN201410182321.8A CN201410182321A CN105095238B CN 105095238 B CN105095238 B CN 105095238B CN 201410182321 A CN201410182321 A CN 201410182321A CN 105095238 B CN105095238 B CN 105095238B
- Authority
- CN
- China
- Prior art keywords
- decision tree
- sample data
- attribute
- transaction
- fraudulent trading
- 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
Landscapes
- Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)
Abstract
The present invention relates to a kind of for detecting the decision tree generation method of fraudulent trading, includes the following steps: to be sampled to form sample data set history fraudulent trading record and history arm's length dealing record;Each element is concentrated to extract the characteristic values of multiple attributes respectively for sample data, attribute includes at least the attribute of the relationship between the current transaction and upper transaction that indicate same transaction card number;Decision tree is constructed with training sample data;Training sample data are divided based on characteristic value, divide decision tree gradually to form decision tree detection model;Carry out test decision tree detection model with test sample data.It considers the correlation between the identical forward and backward transaction of card number, to be easier to detect the fraudulent trading with correlation, and effectively prevent erroneous detection fraudulent trading too much and missing inspection fraudulent trading.
Description
Technical field
The present invention relates to Research on transaction security in electronic fields, more specifically to a kind of for detecting the decision of fraudulent trading
Set generation method and a kind of method based on decision tree detection fraudulent trading.
Background technique
Currently, with the universal of bank card with traded by network prevailing, fraudulent trading is also increasingly multiple, if
It is improper to take precautions against, and can bring loss, or even impact to financial security.Accurately detection fraudulent trading becomes transaction
One of most important R&D direction of security technology area.
In the detection of traditional fraudulent trading, under experimental conditions, can often it be obtained using Decision Tree Algorithm
Good effect, it is good to fraudulent trading detection accuracy, but once apply it under working condition, it is handed over true production
Easy data are tested, then can detect excessive fraudulent trading, precision substantially reduces.To find out its cause, mainly there is following two points:
One, when training pattern, the ratio of 1:1 is often taken in fraudulent trading and arm's length dealing, and under working condition, fraud is handed over
Incident probability only has ten thousand/and it is several, therefore test the data of environment truly restore creation data, this test
Trained model can only be effective to test data under environment, invalid to creation data;If fraudulent trading and arm's length dealing
Ratio using the ratio under production environment, then since the ratio of fraudulent trading and arm's length dealing is excessively greatly different, fraudulent trading
Behavioural characteristic is covered by arm's length dealing completely, and the detection accuracy of model is then too low.
Two, decision-tree model is only capable of carrying out judgement classification to a transaction in isolation.And time transaction before and after same card number
Between usually have certain correlation.For example, there is continuous bankcard consumption until card brush is quick-fried in a short time in a card number
Behavior, then the event may be then pseudo- card fraud.Since each transaction in these transaction may meet normal friendship
These transaction then may be all classified as arm's length dealing by easy condition, decision-tree model, therefore in this case, and decision tree classification is calculated
Rule can fail.
Therefore, this field research staff expectation obtains a kind of for detecting the decision tree generation method of fraudulent trading, with it
The decision tree of generation can overcome when fraudulent trading detection drawbacks described above, more accurate and reliable.
Summary of the invention
The purpose of the present invention is to provide a kind of for detecting the decision tree generation method of fraudulent trading.
To achieve the above object, it is as follows to provide a kind of technical solution by the present invention:
It is a kind of for detecting the decision tree generation method of fraudulent trading, include the following steps: to remember a), to history fraudulent trading
Record and history arm's length dealing record are sampled to form sample data set;B), each element point is concentrated for sample data
The characteristic value of multiple attributes is indescribably taken, attribute includes at least between the current transaction and upper transaction that indicate same transaction card number
Relationship attribute;C), decision tree is constructed using the first part of sample data set as training sample data;D), it is based on feature
Value divides training sample data, divides decision tree gradually to form decision tree detection model;E), with sample data set
Second part carrys out test decision tree detection model as test sample data.
Preferably, step d) specifically comprises the following steps: d1), training sample data are set as to decision tree work as frontal lobe
Node;D2), it is directed to each attribute, current leaf node is divided respectively with multiple mutually different characteristic threshold values, and seeks
Corresponding Gini coefficient after dividing every time;Wherein, characteristic threshold value is in the range of corresponding to the characteristic value of element of the attribute
Any value;D3), to divide the current leaf node with division corresponding to minimum Gini coefficient, to be formed under decision tree
One layer of leaf node;D4), next layer of leaf node is set as current leaf node, repeats step d2), d3), stop until meeting decision tree
Only splitting condition.
Preferably, sample data concentrates the element from history fraudulent trading record to remember with from history arm's length dealing
Ratio of number of the ratio of number of the element of record much smaller than 1 and much larger than history fraudulent trading record and arm's length dealing record.
Preferably, step b) is specifically included: each element concentrated for sample data extracts transaction bank card card number respectively
The characteristic value of attribute, exchange hour attribute and trade company's code attribute of trading;Transaction bank card is pressed to the element that sample data is concentrated
The characteristic value of card number attribute, exchange hour attribute and trade company's code attribute of trading is ranked up;It is concentrated for sample data each
The attribute for the relationship that element addition corresponds between the current transaction and upper transaction of same transaction card number simultaneously extracts the attribute
Characteristic value, wherein relationship include it is same transaction card number current transaction and upper transaction transaction whether having the same trade company
Exchange hour between code and/or the current transaction and upper transaction is poor;It is right for each element that sample data is concentrated
The characteristic value for belonging to continuous variable of the element carries out sliding-model control;The element concentrated to sample data is arranged at random
Sequence.
The present invention also provides a kind of methods based on decision tree detection fraudulent trading, include the following steps: based on above
Decision tree generation method forms decision tree detection model;It is traded with decision tree detection model to production and carries out fraud detection.
It is provided by the present invention for detecting the decision tree generation method of fraudulent trading, overcome in the prior art using determining
The drawbacks of detecting in isolation to single transaction during the detection fraudulent trading of plan tree-model, the forward and backward pen of identical card number is handed over
For the attribute of relationship between easily as one in multiple attributes of sample data concentration element, this data processing method can be with
The correlation between the identical forward and backward transaction of card number is considered during training pattern, to be easier there will be correlation
Fraudulent trading detection;In addition, present invention improves over the ratios of fraudulent trading sample during model training and arm's length dealing sample
Example choose, so that the recall rate of model and detection accuracy is fallen on reasonable section, effectively prevent erroneous detection fraudulent trading too much and
Missing inspection fraudulent trading.
Detailed description of the invention
Fig. 1 shows the flow chart of the decision tree generation method for detecting fraudulent trading of one embodiment of the invention offer.
Step S13 in the decision tree generation method for detecting fraudulent trading provided Fig. 2 shows one embodiment of the invention
Execution flow chart.
Specific embodiment
As shown in Figure 1, one embodiment of the invention provides a kind of for detecting the decision tree generation method of fraudulent trading, packet
Include following steps:
Step S10, history fraudulent trading record and history arm's length dealing record are sampled to form sample data
Collection.
Specifically, all fraudulent trading data in history fraudulent trading table are extracted, card number is pressed from history arm's length dealing table
Arm's length dealing data are extracted, fraudulent trading, arm's length dealing can be respectively labeled as fraud, normal.If only historical trading table and
History fraudulent trading table (does not separate history arm's length dealing table), then due to including fraudulent trading data in historical trading table, is taking out
When taking arm's length dealing data, need will be comprising taking out in historical trading table again after the record in fraudulent trading table is rejected
It takes.Sample data set is formed after extracting proper data.
Due to that may include mass data in historical trading table, only need to therefrom sample out partial data in training decision tree
?.Sample mode can be random sampling.
Further, sample data is concentrated from the element of history fraudulent trading record and from history arm's length dealing
Ratio of number of the ratio of number of the element of record much smaller than 1 and much larger than history fraudulent trading record and arm's length dealing record.
The ratio can make while not covering fraudulent trading feature, be closer in creation data fraudulent trading with just
The actual proportions often traded.
For example, being drafted from the element of history fraudulent trading record and from history just during data from the sample survey
It, can be with reference to the actual proportions of fraudulent trading and arm's length dealing in creation data when the ratio of number of the element of normal transaction record
(actual proportions are, for example, 1:10000), and on its basis reduce the 1-2 order of magnitude, such as can value be 1:100.
Step S11, each element is concentrated to extract the characteristic values of multiple attributes respectively for sample data.
Wherein, attribute includes at least the category of the relationship between the current transaction and upper transaction that indicate same transaction card number
Property.
Specifically, element includes multiple primitive attributes, such as: trade date, exchange hour, card property etc..It is same to consider
Correlation before and after one card number between time transaction is that element adds context property according to above-mentioned primitive attribute, for example, selected member
Plain attribute is as shown in table 1 below:
Table 1
Wherein, primitive attribute is directly acquired from fraudulent trading table and historical trading table, does not need to be calculated;On
Hereafter attribute needs to need to carry out certain calculating or judgement from obtaining in the data of transaction on same card number.Exchange hour
Poor attribute (time_diff) indicates the time difference on same card number between transaction and current transaction, if same trade company's attribute
(is_same_mchnt) indicate that whether transaction and current transaction are in the generation of same trade company on same card number.
The decision tree generation method for being used to detect fraudulent trading that an embodiment provides according to the present invention, for each element point
The specific execution process for the step of indescribably taking the characteristic value of multiple attributes is as follows:
The each element concentrated for sample data extracts transaction bank card card number attribute, exchange hour attribute and transaction respectively
The characteristic value of trade company's code attribute;
Transaction bank card card number attribute, exchange hour attribute and transaction trade company's code category are pressed to the element that sample data is concentrated
The characteristic value of property is ranked up;
Correspond to the current transaction and upper transaction of same transaction card number for each element addition that sample data is concentrated
Between relationship attribute and extract the characteristic value of the attribute, wherein relationship include same transaction card number current transaction with it is upper
Whether having the same exchange hour between transaction transaction trade company's code and/or the current transaction and upper transaction be poor;
For each element that sample data is concentrated, discretization is carried out to the characteristic value for belonging to continuous variable of the element
Processing;
The element that sample data is concentrated is carried out randomly ordered.
For example, carrying out the corresponding relationship that sliding-model control is referred to the following table 2 to the characteristic value of exchange hour difference attribute:
Value before exchange hour difference discretization | Value after exchange hour difference discretization |
Time_diff≤1 minute | 1 |
1 minute < time_diff≤10 minute | 2 |
10 minutes < time_diff≤30 minute | 3 |
30 minutes < time_diff≤1 hour | 4 |
1 hour < time_diff≤6 hour | 5 |
6 hours < time_diff≤1 day | 6 |
1 day < time_diff≤7 day | 7 |
7 days < time_diff≤1 month | 8 |
1 month < time_diff≤6 month | 9 |
Time_diff > 6 month | 10 |
Table 2
Step S12, decision tree is constructed using the first part of sample data set as training sample data.
Specifically, such as the data of extraction sample data concentration 2/3rds are as training sample data, as most early years
Node constructs decision tree.Sample data set is gradually divided into multiple sons and gradually dividing decision tree in the next steps
Each elemental recognition in sample data set is finally arm's length dealing or fraudulent trading by collection.
Step S13, training sample data are divided based on characteristic value, divides decision tree gradually to form decision tree inspection
Survey model.
Fig. 2 shows the specific execution processes of step S13 comprising as follows step by step:
Step S130, training sample data are set as to the current leaf node of decision tree.
Specifically, which is the initialization step to form decision tree detection model.
Step S131, it is directed to each attribute, current leaf node is drawn respectively with multiple mutually different characteristic threshold values
Point, and seek Gini coefficient corresponding after dividing every time.
Wherein, characteristic threshold value any value in the range of corresponding to the characteristic value of element of the attribute.More specifically, special
Levy any value in the range of the characteristic value of element of the threshold value included by current leaf node.
For example, it is assumed that the element that current leaf node includes is set E:{ e1,e2,e3,…en, element has altogether in upper table 1
The 18 attribute { X shown1,X2,X3,…X18, for any attribute Xi(1≤i≤18), if each element { e1,e2,e3,…en}
Attribute Xi characteristic value formed set C:{ ci1,ci2,ci3,…,cim, then with the either element C in set Cir(1≤r≤m)
Current leaf node is divided, the feature of attribute Xi as characteristic threshold value (i.e. characteristic threshold value can arbitrarily be chosen in set C)
Value is less than or equal to CirCurrent leaf node element formed subset T(Xi≤Cir), the characteristic value of attribute Xi is greater than CirWork as frontal lobe
The element of node forms subset T(Xi > Cir), that is, set E is divided into subset T(Xi≤Cir) and subset T(Xi > Cir).
Gini coefficient is sought to the secondary division, its calculation formula is:
Wherein, c is the characteristic threshold value for any attribute Xi, and Xi is the ith attribute that sample data concentrates element, Gini
(TXi=c) it is corresponding Gini coefficient after threshold value divides current leaf node characterized by c, T (Xi≤c) is after dividing
Subset composed by element of the characteristic value of attribute Xi less than or equal to c, T (Xi > c) are that the characteristic value of attribute Xi after dividing is greater than c
Element composed by subset, Num (T (Xi≤c)) be subset T (Xi≤c) in element quantity, Num (T (X > c)) be subset T
The quantity of element in (Xi > c), Gini (T (Xi≤c)) are the Gini coefficient of subset T (Xi≤c), and Gini (T (Xi > c)) is subset
The Gini coefficient of T (Xi > c).
Element set E:{ e1,e2,e3,…enSubset T Gini coefficient by following equation group calculate:
,
Wherein, pnormal(T) probability of the element recorded from history arm's length dealing in subset T, p are indicatedfraud(T)
Indicate probability of the element recorded from history fraudulent trading in subset T.
Specifically, in this step by step S131, for any attribute Xi, with set C:{ ci1,ci2,ci3,…,cimIn
Each element divides current leaf node respectively as characteristic threshold value, acquires corresponding Gini coefficient Gini after dividing every time
(TXi=c) and record.Another attribute Xj of reselection (1≤j≤18 and j ≠ i) carries out same operation, until to all 18
A attribute all completes same operation.
Step S132, to divide the current leaf node with division corresponding to minimum Gini coefficient, to form decision tree
Next layer of leaf node.
Theoretical, corresponding Gini coefficient after more above-mentioned multiple division, choosing are divided according to the decision tree based on Gini coefficient
Take the smallest best divisional mode for being divided into decision tree of Gini coefficient.Work as frontal lobe with the best divisional mode division decision tree
Node forms the next layer of leaf node (i.e. two subsets) of decision tree.
Step S133, judge whether that meeting decision tree stops splitting condition, if satisfied, then the process of step S13 terminates, it is no
Then, the following steps S134 is executed.
Specifically, it includes either one or two of following condition that decision tree, which stops splitting condition:
Condition 1: subset T (Xi≤c), subset T (Xi > c) are normal from history fraudulent trading record or history in the same manner
Transaction record;
Condition 2: corresponding Gini coefficient is more than or equal to Gini coefficient corresponding after preceding primary division after current division;
Condition 3: number of elements included by subset T (Xi≤c) or subset T (Xi > c) is less than number of elements threshold value.
Step S134, next layer of leaf node is set as current leaf node, repeats step S131, S132.
Specifically, in the case where not meeting above-mentioned stopping splitting condition, step S134 persistently carries out decision tree
Division.
Step S14, carry out test decision tree detection model using the second part of sample data set as test sample data.
Specifically, can be concentrated using sample data the data of remaining one third tested as test sample data by
The decision tree detection model formed according to above-described embodiment.
For detecting the decision tree generation method of fraudulent trading provided by the above embodiment of the present invention, existing skill is overcome
The drawbacks of in art using being detected in isolation to single transaction during decision-tree model detection fraudulent trading, by identical card number
The attribute of relationship between forward and backward transaction concentrates one in multiple attributes of element as sample data, at this data
Reason mode can consider the correlation between the identical forward and backward transaction of card number during training pattern, so that being easier will
Fraudulent trading detection with correlation.
Another embodiment of the present invention provides a kind of method based on decision tree detection fraudulent trading, this method includes following step
It is rapid:
A), the decision tree generation method provided based on above embodiments forms decision tree detection model;
B), traded with decision tree detection model to production and carry out fraud detection.
Further, further include following steps after step B): it is artificial true to obtain testing result progress to fraud detection
Recognize, and history fraudulent trading record is added in the record for the generation transaction for being confirmed to be fraudulent trading.Updated history fraud
Transaction record can be used to regenerate decision tree detection model.It will be appreciated by those skilled in the art that in the above manner every one
The section time regenerates decision tree detection model, it is ensured that detection model can identify feature and the rule of newest fraudulent trading
Rule.
Further improved embodiment according to this embodiment, when decision tree generation method in implementation steps A),
Following steps can also be performed after step S14:
If recall rate is greater than first threshold, history fraudulent trading record and history arm's length dealing record are taken out again
Sample, to correspondingly improve the ratio for deriving from the element of history fraudulent trading record and concentrating in sample data;If recall rate is small
In second threshold, then samples to history fraudulent trading record and history arm's length dealing record, derived from accordingly decreasing again
The ratio that the element of history fraudulent trading record is concentrated in sample data;
Continue to execute decision tree generation method corresponding step S11, S12, S13 and S14;
Wherein, recall rate is to be detected in number of elements and test sample data for fraudulent trading in test sample data
From the ratio between the number of elements of history fraudulent trading record, first threshold is greater than second threshold.Recall rate is high, indicates detection mould
Type erroneous detection situation is more;Recall rate is low, indicates that detection model missing inspection situation is more;Recall rate will all indicate to detect without falling into zone of reasonableness
Precision is undesirable.
Therefore, invention also improves the ratios of fraudulent trading sample and arm's length dealing sample during model training to select
It takes, the recall rate of model and detection accuracy is made to fall on reasonable section, effectively prevent erroneous detection fraudulent trading and missing inspection too much
Fraudulent trading.
Above description is not lain in and is limited the scope of the invention only in the preferred embodiment of the present invention.Ability
Field technique personnel can make various modifications design, without departing from thought of the invention and subsidiary claim.
Claims (8)
1. it is a kind of for detecting the decision tree generation method of fraudulent trading, include the following steps:
A), history fraudulent trading record and history arm's length dealing record are sampled to form sample data set;
B), each element is concentrated to extract the characteristic values of multiple attributes respectively for the sample data, the attribute includes at least
It indicates the attribute of the relationship between the current transaction and upper transaction of same transaction card number, transaction bank card card number attribute, hand over
Easy time attribute and transaction trade company's code attribute;
C), decision tree is constructed using the first part of the sample data set as training sample data;
D), the training sample data are divided based on the characteristic value, divides the decision tree gradually to form decision
Set detection model;And
E), the decision tree detection model is tested as test sample data using the second part of the sample data set;
Wherein, the step b) is specifically included:
For the sample data concentrate each element extract respectively the transaction bank card card number attribute, exchange hour attribute and
The characteristic value for trade company's code attribute of trading;
In the transaction bank card card number attribute, exchange hour attribute and transaction trade company's generation, are pressed to the element that the sample data is concentrated
The characteristic value of code attribute is ranked up;
Correspond to the current transaction and upper transaction of same transaction card number for each element addition that the sample data is concentrated
Between relationship attribute and extract the characteristic value of the attribute, wherein the relationship includes the current transaction of same transaction card number
With the exchange hour between upper transaction transaction trade company's code whether having the same and/or the current transaction and upper transaction
Difference;
For each element that the sample data is concentrated, discretization is carried out to the characteristic value for belonging to continuous variable of the element
Processing;
The element that the sample data is concentrated is carried out randomly ordered.
2. decision tree generation method according to claim 1, which is characterized in that the step d) specifically includes following step
It is rapid:
D1), the training sample data are set as to the current leaf node of the decision tree;
D2), it is directed to each attribute, the current leaf node is drawn respectively with multiple mutually different characteristic threshold values
Point, and seek Gini coefficient corresponding after dividing every time;Wherein, the characteristic threshold value is in the element for corresponding to the attribute
Characteristic value in the range of any value;
D3), to divide the current leaf node with division corresponding to the minimum Gini coefficient, to form the decision tree
Next layer of leaf node;
D4), the next layer of leaf node is set as the current leaf node, repeating said steps d2), d3), until meet decision
Tree stops splitting condition.
3. decision tree generation method according to claim 2, which is characterized in that the calculation formula of the Gini coefficient are as follows:
Wherein, c is the characteristic threshold value for any attribute, and Xi is the ith attribute that the sample data concentrates element,
Gini(TXi=c) be take c as Gini coefficient corresponding after the characteristic threshold value divides the current leaf node, T (Xi
≤ c) it is subset composed by element of the characteristic value less than or equal to c of attribute Xi after dividing, T (Xi > c) is attribute Xi after dividing
Subset composed by element of the characteristic value greater than c, Num (T (Xi≤c)) are the quantity of element in subset T (Xi≤c), Num (T (X
> c)) be subset T (Xi > c) in element quantity, Gini (T (Xi≤c)) be subset T (Xi≤c) Gini coefficient, Gini (T
(Xi > c)) be subset T (Xi > c) Gini coefficient.
4. decision tree generation method according to claim 3, which is characterized in that the decision tree stops splitting condition and includes
Either one or two of following condition:
The subset T (Xi≤c), subset T (Xi > c) are deriving from history fraudulent trading record or the history just in the same manner
Normal transaction record;
Corresponding Gini coefficient is more than or equal to Gini coefficient corresponding after preceding primary division after current division;
Number of elements included by the subset T (Xi≤c) or subset T (Xi > c) is less than number of elements threshold value.
5. decision tree generation method according to claim 1, which is characterized in that the sample data is concentrated from described
The element of history fraudulent trading record is remote much smaller than 1 with the ratio of number of the element from history arm's length dealing record
Greater than the ratio of number of history fraudulent trading record and arm's length dealing record.
6. a kind of method based on decision tree detection fraudulent trading, includes the following steps:
Based on decision tree generation method described in any one of claims 1 to 5, the decision tree detection model is formed;
Fraud detection is carried out to transaction is generated with the decision tree detection model.
7. the method for detection fraudulent trading according to claim 6, which is characterized in that it further includes following steps:
Testing result is obtained to the fraud detection and carries out manual confirmation, and the generation for being confirmed to be fraudulent trading is traded
Record history fraudulent trading record is added.
8. it is according to claim 7 detection fraudulent trading method, which is characterized in that in the decision tree generation method
Further include following steps after the step e):
If recall rate is greater than first threshold, history fraudulent trading record and the history arm's length dealing are recorded again
Sampling, to correspondingly improve the ratio for deriving from the element of history fraudulent trading record and concentrating in the sample data;Such as
Recall rate described in fruit is less than second threshold, then takes out again to history fraudulent trading record and history arm's length dealing record
Sample, to accordingly decrease the ratio for deriving from the element of history fraudulent trading record and concentrating in the sample data;
Continue to execute the step b), c), d) and e) of the decision tree generation method;
Wherein, the recall rate is the number of elements and the test specimens being detected in the test sample data as fraudulent trading
From the ratio between the number of elements of history fraudulent trading record in notebook data, the first threshold is greater than second threshold
Value.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410182321.8A CN105095238B (en) | 2014-05-04 | 2014-05-04 | For detecting the decision tree generation method of fraudulent trading |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410182321.8A CN105095238B (en) | 2014-05-04 | 2014-05-04 | For detecting the decision tree generation method of fraudulent trading |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105095238A CN105095238A (en) | 2015-11-25 |
CN105095238B true CN105095238B (en) | 2019-01-18 |
Family
ID=54575698
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410182321.8A Active CN105095238B (en) | 2014-05-04 | 2014-05-04 | For detecting the decision tree generation method of fraudulent trading |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105095238B (en) |
Families Citing this family (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106844367A (en) * | 2015-12-03 | 2017-06-13 | 阿里巴巴集团控股有限公司 | A kind of processing method and processing device of Internet service |
CN106897931A (en) * | 2016-06-12 | 2017-06-27 | 阿里巴巴集团控股有限公司 | A kind of recognition methods of abnormal transaction data and device |
CN107657453B (en) * | 2016-07-25 | 2020-10-20 | 平安科技(深圳)有限公司 | Method and device for identifying fraudulent data |
CN106548343B (en) * | 2016-10-21 | 2020-11-10 | 中国银联股份有限公司 | Illegal transaction detection method and device |
CN106529960A (en) * | 2016-11-07 | 2017-03-22 | 中国银联股份有限公司 | Fraud transaction detection method for electronic transaction |
CN106548302B (en) * | 2016-12-01 | 2020-08-14 | 携程旅游信息技术(上海)有限公司 | Risk identification method and system for internet transaction |
CN107679862B (en) * | 2017-09-08 | 2021-08-27 | 中国银联股份有限公司 | Method and device for determining characteristic value of fraud transaction model |
CN110298663B (en) * | 2018-03-22 | 2023-04-28 | 中国银联股份有限公司 | Fraud transaction detection method based on sequence wide and deep learning |
CN108334647A (en) * | 2018-04-12 | 2018-07-27 | 阿里巴巴集团控股有限公司 | Data processing method, device, equipment and the server of Insurance Fraud identification |
CN109034194B (en) * | 2018-06-20 | 2022-03-04 | 东华大学 | Transaction fraud behavior deep detection method based on feature differentiation |
CN108805580A (en) * | 2018-06-21 | 2018-11-13 | 上海银赛计算机科技有限公司 | Account number analysis method, device and storage medium |
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 |
CN110874479B (en) * | 2018-08-29 | 2023-05-16 | 创新先进技术有限公司 | Method, system, data terminal and processing terminal for safely processing decision tree model |
US11321632B2 (en) * | 2018-11-21 | 2022-05-03 | Paypal, Inc. | Machine learning based on post-transaction data |
CN110706090A (en) * | 2019-08-26 | 2020-01-17 | 阿里巴巴集团控股有限公司 | Credit fraud identification method and device, electronic equipment and storage medium |
CN110717509B (en) * | 2019-09-03 | 2024-04-05 | 中国平安人寿保险股份有限公司 | Data sample analysis method and device based on tree splitting algorithm |
US11551230B2 (en) * | 2020-01-14 | 2023-01-10 | Visa International Service Association | Security attack detections for transactions in electronic payment processing networks |
CN111415167B (en) * | 2020-02-19 | 2023-05-16 | 同济大学 | Network fraud transaction detection method and device, computer storage medium and terminal |
CN111507382B (en) * | 2020-04-01 | 2023-05-05 | 北京互金新融科技有限公司 | Sample file clustering method and device and electronic equipment |
CN112070107A (en) * | 2020-07-15 | 2020-12-11 | 上海大学 | Electronic port ship harboring control method |
CN112801231B (en) * | 2021-04-07 | 2021-07-06 | 支付宝(杭州)信息技术有限公司 | Decision model training method and device for business object classification |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110251951A1 (en) * | 2010-04-13 | 2011-10-13 | Dan Kolkowitz | Anti-fraud event correlation |
US8726379B1 (en) * | 2011-07-15 | 2014-05-13 | Norse Corporation | Systems and methods for dynamic protection from electronic attacks |
CN102890803B (en) * | 2011-07-21 | 2016-01-06 | 阿里巴巴集团控股有限公司 | The defining method of the abnormal process of exchange of electronic goods and device thereof |
CN103544429B (en) * | 2012-07-12 | 2016-12-21 | 中国银联股份有限公司 | The abnormal detector mutual for safety information and method |
CN103678659A (en) * | 2013-12-24 | 2014-03-26 | 焦点科技股份有限公司 | E-commerce website cheat user identification method and system based on random forest algorithm |
-
2014
- 2014-05-04 CN CN201410182321.8A patent/CN105095238B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN105095238A (en) | 2015-11-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105095238B (en) | For detecting the decision tree generation method of fraudulent trading | |
WO2022110557A1 (en) | Method and device for diagnosing user-transformer relationship anomaly in transformer area | |
CN104679777B (en) | A kind of method and system for being used to detect fraudulent trading | |
WO2017143932A1 (en) | Fraudulent transaction detection method based on sample clustering | |
CN106991447A (en) | A kind of embedded multi-class attribute tags dynamic feature selection algorithm | |
CN105975959A (en) | Face characteristic extraction modeling method based on neural network, face identification method, face characteristic extraction modeling device and face identification device | |
CN104933443A (en) | Automatic identifying and classifying method for sensitive data | |
TW201732662A (en) | Method and apparatus for establishing data identification model | |
CN106326913A (en) | Money laundering account determination method and device | |
CN109903053B (en) | Anti-fraud method for behavior recognition based on sensor data | |
CN109727446A (en) | A kind of identification and processing method of electricity consumption data exceptional value | |
CN111539450B (en) | Landslide susceptibility evaluation method in earthquake region | |
CN103473786A (en) | Gray level image segmentation method based on multi-objective fuzzy clustering | |
CN104091078B (en) | Product Multi-information acquisition indicating failure means to save the situation based on D S evidence theories | |
CN108268886B (en) | Method and system for identifying plug-in operation | |
CN104850868A (en) | Customer segmentation method based on k-means and neural network cluster | |
Wang et al. | Mining changes of classification by correspondence tracing | |
CN106228190A (en) | Decision tree method of discrimination for resident's exception water | |
CN106971344A (en) | Insured amount control method and system | |
CN109255029A (en) | A method of automatic Bug report distribution is enhanced using weighted optimization training set | |
CN115148299A (en) | XGboost-based ore deposit type identification method and system | |
CN102521534A (en) | Intrusion detection method based on crude entropy property reduction | |
CN103902798B (en) | Data preprocessing method | |
Wu et al. | Track-clustering error evaluation for track-based multi-camera tracking system employing human re-identification | |
CN108229011A (en) | A kind of shale lithofacies development Dominated Factors judgment method, equipment and storage device |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |