CN106372655A - Synthetic method for minority class samples in non-balanced IPTV data set - Google Patents
Synthetic method for minority class samples in non-balanced IPTV data set Download PDFInfo
- Publication number
- CN106372655A CN106372655A CN201610753263.9A CN201610753263A CN106372655A CN 106372655 A CN106372655 A CN 106372655A CN 201610753263 A CN201610753263 A CN 201610753263A CN 106372655 A CN106372655 A CN 106372655A
- Authority
- CN
- China
- Prior art keywords
- sample
- minority class
- samples
- dangerous
- major
- 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.)
- Pending
Links
Classifications
-
- 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
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a synthetic method for minority class samples in a non-balanced IPTV data set, and aims to overcome the defect of performance reduction of a subsequent classification and prediction model caused by the fact that new samples are directly generated without analytic processing of minority samples in an existing minority class data synthesis method. The synthetic method is implemented by the steps of firstly finding out a neighbor set of the minority class samples; dividing neighbor samples into a noise set, a security set and a dangerous set according to a proportion of categories which the neighbor samples belong to; not processing samples in the noise set; calculating a ratio of the security set to the dangerous set, and calculating a related probability; selecting the security set or the dangerous set according to the probability; and generating new minority class samples based on the samples in the selected set. By adopting the method, the minority class sample effect having the negative effect for classification can be removed; the utility of the minority class samples near a classification face is improved; and the obtained new minority class samples can better improve the performance of the subsequent classification and prediction model.
Description
Technical field
The present invention relates to non-equilibrium data process field, especially relate to the minority class on a kind of non-equilibrium iptv data set
The synthetic method of sample.
Background technology
With the business transformation of domestic fixed network operator, become the new industry of operator based on the various value-added services of the Internet
The important component part of business growth point, especially IPTV (iptv) business has presented the situation of rapid growth.
Iptv has following features: (1) user is obtained in that high-quality digital media service；(2) user can pass through broadband ip network
Free selection video frequency program；(3) it provides wide emerging market for operator.In recent years, operator and research institution
Personnel are devoted to lifting impression and the satisfaction of iptv user by the key factor of research impact user experience quality (qoe)
Degree.
In existing solution, the report based on the status data gathering from iptv Set Top Box and user hinders data, leads to
Cross the model in machine learning and correlation technique to predict the qoe of user.But due in iptv business in most cases, net
In order, Consumer's Experience is also preferable, does not report barrier for network, in limited instances poor user experience and report barrier, thus Set Top Box institute
The data collecting is nonequilibrium, i.e. there are two class users report barrier classifications and user does not report barrier classification.Wherein use
The sample number of family report barrier classification is far smaller than the sample number that user does not report barrier classification, then in this problem, user's report barrier classification
For minority class, it is many several classes ofs that user does not report barrier classification.
In order to solve non-equilibrium data process problem it is often necessary to according to available data characteristic, synthesize a part of minority class
Sample, so that two class data volumes reach balance.In existing method, synthetic minority
Oversampling technique (smote), as the technology of an over-sampling, is frequently utilized for synthesizing minority class.Although
Smote algorithm has many good qualities, but still has some defects, including over-fitting data polytropy.Particularly, work as smote
Generate equal number of generated data for each a few sample, neighbours' sample is not taken into account, this can increase minority class
The probability that internal specimen overlapping phenomenon occurs.In addition some minority class samples are located near classification interface, and subsequent classifier is risen
Pivotal role, and other samples are located at most apoplexy due to endogenous wind, belong to noise, generate minority class sample if based on it, then can be right
Classification has the opposite effect, and existing smote algorithm does not consider these problems.Based on this, the present invention specifically addresses smote technology
Some technological deficiencies existing, preferably solve the problems, such as the data nonbalance in iptv user qoe prediction.
Content of the invention
The technical problem to be solved is that the deficiency for background technology provides a kind of non-equilibrium iptv data
The synthetic method of the minority class sample on collection.
The present invention is to solve above-mentioned technical problem to employ the following technical solutions:
A kind of synthetic method of the minority class sample on non-equilibrium iptv data set, specifically includes following steps:
Step 1: find out minority class sample set xminorIn each sample point xiCorresponding k neighbour set si, wherein k is nature
Number, i=1 ... n, xi∈xminor；K neighbour collection is combined into apart from xiThe set that k nearest sample is formed；
Step 2, the characteristic of k each the minority class sample of neighbour's set analysis being obtained according to step 1, and then be classified as making an uproar
Sound collection, safe collection and dangerous collection three classes；
Step 3, the sample that noise is concentrated does not process, and calculates the sample size in safe collection and the dangerous sample concentrated
Ratio t between quantity；
Step 4, produces an equally distributed random number b obeying on interval [0,1]；If b is ∈ [0, t/ (t+1)], then
Select the dangerous all samples concentrated as input, the smote algorithm sending into standard generates new minority class sample；Conversely, then
Select all samples in safe collection as input, the smote algorithm sending into standard generates new minority class sample；
Step 5, original minority class sample and newly-generated minority class sample are combined the new minority class set of composition
Close.
Further preferred side as the synthetic method of the minority class sample on a kind of non-equilibrium iptv data set of the present invention
Case, described step 2 specifically comprises the steps of:
Step 2.1, counts siIn belong to many several classes ofs xmajorNumber of samples, use | si∩xmajor| to represent, its expression
Many several classes ofs sample set xmajorAnd siCommon factor in number of samples.
Step 2.2, judges | si∩xmajor| residing interval, it is specifically divided into three kinds of situations:
If | si∩xmajor|=k, then current sample xiIt is in most apoplexy due to endogenous wind, it is believed that it is to make an uproar for classification problem
Sound；xminorIn all samples composition safe collections meeting this condition；
If 0≤| si∩xmajor| < 0.5k then shows current sample xiDangerous very little by misclassification；xminorIn all full
The sample composition safe collection of this condition of foot；
If 0.5k≤| si∩xmajor| < k then shows current sample xiExist by the danger of misclassification；xminorIn all full
The dangerous collection of sample composition of this condition of foot.
Further preferred side as the synthetic method of the minority class sample on a kind of non-equilibrium iptv data set of the present invention
Case, in step 4, the concrete calculating process of algorithm of described smote is as follows: sets current sample as xi, from the k neighbour of this sample
Set siOne sample x of middle random selectionj, produce an equally distributed random number δ of obedience from interval [0,1], then newly-generated
Minority class sample is: xnew=xi+δ×(xj-xi).
The present invention adopts above technical scheme compared with prior art, has following technical effect that
1. the present invention can solve the problems such as classification of non-equilibrium data, prediction by producing minority class sample；
2. the present invention classifies to minority class sample, does not consider using the minority class sample being absorbed among many several classes ofs sample
Produce new samples, it is to avoid the hydraulic performance decline being brought in subsequent classification by noise.Further, since the dangerous sample concentrated is in two
Near the classification interface of class, the minority class sample new using the sample generation in this set as much as possible, be conducive to significantly
Improve subsequent classification, the performance of Forecasting Methodology；
3. the data overlap during the present invention can avoid the minority class sample that traditional smote algorithm is brought to produce
Problem.
Brief description
Fig. 1 is the synthetic method flow chart of the minority class sample on the present invention non-equilibrium iptv data set；
Fig. 2 is to be respectively adopted three kinds of methods under knn grader of the present invention to process the g average ratio of non-equilibrium iptv data sets relatively
Result；
Fig. 3 is to be respectively adopted the g average ratio that three kinds of methods process non-equilibrium iptv data set under c4.5 grader of the present invention
Relatively result；
Fig. 4 is that the present invention is respectively adopted the minority class data that the smote method of standard and method proposed by the present invention generate
G average comparative result as test set.
Specific embodiment
Below in conjunction with the accompanying drawings technical scheme is described in further detail:
As shown in figure 1, a kind of synthetic method of the minority class sample on non-equilibrium iptv data set, its step includes:
Step 1: find out all minority class sample points respective k neighbour set si, wherein k is natural number, and i is positive integer；
Step 2, the characteristic of k each the minority class sample of neighbour's set analysis being obtained according to step 1, and then be classified as making an uproar
Sound collection, safe collection and dangerous collection three classes；
Step 3, the sample that noise is concentrated does not process, and calculates the sample size in safe collection and the dangerous sample concentrated
Ratio t between quantity；
Step 4, produces an equally distributed random number b obeying on interval [0,1]；If b is ∈ [0, t/ (t+1)], then
Select the dangerous all samples concentrated as input, the smote algorithm sending into standard generates new minority class sample；Conversely, then
Select all samples in safe collection as input, the smote algorithm sending into standard generates new minority class sample；
Step 5, original minority class sample and newly-generated minority class sample are combined the new minority class set of composition
Close.
The detailed process of all steps is as follows:
Step 1: set the data that iptv Set Top Box collects and include status dataHinder data with the report of userBoth are one-to-one.Wherein vector xiDimension be p, reflect iptv network condition (time delay, packet loss, interim card
Deng), yiFor scalar, it is the labelling whether user reports barrier, such as user ensures, then yi=1, conversely, yi=0.So, minority class sample
This collection xminorIt is defined as yi=1, i=1 ..., corresponding all x of ni；Many several classes ofs sample set xmajorIt is defined as yi=0, i
=1 ..., corresponding all x of ni, i.e. xmajor=x xmajor.For each sample x in minority classi∈xminor, calculate
Its with x in all samples Euclidean distance, k nearest sample of selected distance form xiK neighbour set si.
Step 2: by the characteristic of k each minority class sample of neighbour's set analysis, minority class sample is classified further, specifically
As follows:
(2-1) count siIn k sample in belong to many several classes ofs xmajorNumber of samples, that is, obtain | si∩xmajor|,
This can be by counting siMiddle sample generic labelling y obtains.
(2-2) judge | si∩xmajor| residing interval, it is divided into three kinds of situations:
Situation 1: if | si∩xmajor|=k, then show current sample xiIt is in most apoplexy due to endogenous wind, for classification problem
Speech is it is believed that it is noise.xminorIn the set of all samples compositions meeting this condition be defined as " safe collection "；
Situation 2: if 0≤| si∩xmajor| < 0.5k then shows current sample xiDangerous very little by misclassification.xminor
In the set of all samples compositions meeting this condition be defined as " safe collection "；
Situation 3: if 0.5k≤| si∩xmajor| < k then shows current sample xiExist by the danger of misclassification.xminor
In the set of all samples compositions meeting this condition be defined as " dangerous collection "；
(2-3) sample point concentrated for the noise meeting situation 1, it does not do any subsequent treatment, i.e. do not utilize its life
The minority class sample of Cheng Xin.For the dangerous sample point concentrated of the safe collection meeting situation 2 and situation 3, enter next step
Continue with.
Step 3: calculate the ratio between the sample size in safe collection obtained in the previous step and the dangerous sample size concentrated
Value, is designated as t.
Step 4: produce an equally distributed random number obeyed on interval [0,1], be designated as b.If b is ∈ [0, t/ (t+
1)], then with the dangerous all samples concentrated as input, send into standard smote algorithm and generate new minority class sample；No
Then, with all samples in safe collection as input, the smote algorithm sending into standard generates new minority class sample.Original few
Several classes of sample and newly-generated minority class sample are combined, and form new minority class set.
Standard smote algorithm in this step is as follows: sets current sample as xi, from the k neighbour set s of this sampleiIn with
Machine selects a sample xj, produce an equally distributed random number δ of obedience from interval [0,1], then newly-generated minority class sample
Originally it is: xnew=xi+δ×(xj-xi).
It should be noted that needing the new sample number producing by between many several classes ofs sample number and original minority class sample number
Difference determine.Assume that many several classes ofs sample and minority class sample size are respectively | xmajor| and | xminor|, then need newly-generated
(xmajor|-|xminor|) individual minority class sample.If this step is with safe collection, and (sample number in this set is nsafe) as standard
The input of smote algorithm, then each sample in safe collection need operation standard smote algorithmSecondary.With
Reason, if the danger of this step integrates, and (sample number in this set is as ndanger) as standard smote algorithm input, then dangerous
Each sample concentrated needs operation standard smote algorithmSecondary.
Embodiment and performance evaluation
In order to the synthetic method of the minority class sample non-equilibrium iptv data set present invention designed by is better described
Advantage, be applied to prediction iptv system user report barrier.Here, two original data sets both are from Jiangsu Telecom.
Data set 1 (i.e. x) is to April iptv Key Performance Indicator (kpi) data of No. ten from April No. one.Data set 2 (i.e. y) is
Hinder data (the user's report barrier data receiving by phone) from the report of user.
After collecting raw data set, need to carry out data cleansing to it, its object is to remove in initial data
Repeat to record, the data such as error logging and property value disappearance record, and by the data data collection 2 in data set 1
Data corresponds, and according to the report barrier labelling of data set 2, the data in data set 1 is classified, for use as subsequently pre-
Survey the training of model.After data cleansing, in data set x, total record (sample) has 439050, wherein 4871 genus
In minority class, 434179 belong to many several classes ofs, and dimension p of each data is 11.The implication of each dimension is to be shown in Table 1.
The implication of each dimension of table 1 data
After data cleansing, for equilibrium majority class and minority class sample, using the non-equilibrium iptv designed by the present invention
The synthetic method of the minority class sample on data set, produces minority class sample so that new minority class sample total is original number
According to the minority class sample number concentrated 40 times.
With several classes of sample more than 150000 and 150000 minority class samples (new) as training dataset, have chosen here
K arest neighbors (knn) sorting algorithm and c4.5 Decision Tree Algorithm implementation model training, and the model logarithm with training
Classified according to the remaining data concentrated.Fig. 2 directly carries out for not producing new minority class sample under knn grader classifying,
Generate new minority class sample only with standard smote method to be classified and using method proposed by the present invention new the lacking of generation
Several classes of sample carries out the comparative result of the g average (g-mean) under three kinds of methods of classifying.
In figs. 2 and 3, the ratio for minority class and many several classes ofs in the data set classified is 1:20 respectively
(6000:12000), 1:25 (6000:15000) and 1:30 (6000:18000).Under the test case of these three ratios, we
As can be seen that either knn grader or c4.5 grader, the g of the minority class sample synthetic method designed by the present invention is equal
Value is higher than other two methods.Longitudinally contrast finds Fig. 2 and Fig. 3, compares with c4.5 grader, knn grader and the present invention
The minority class sample synthetic method being proposed combines, and has more preferable classification performance.
Additionally, the minority class data generating the smote method of standard and method proposed by the present invention, as test set, is come
Compare the g average of two methods.The numeral that Fig. 4 can be seen that on transverse axis represents by standard smote method and present invention proposition
The number of minority class that generates respectively of method.In test data, the ratio of minority class and many several classes ofs is is 1:20.This three
In the case of kind, it may be seen that the g average of method proposed by the present invention is above standard smote method.
Test result indicate that using the minority class sample synthetic method designed by the present invention, significantly improving existing non-equilibrium
The classification estimated performance of iptv data set.
Claims (3)
1. the minority class sample on a kind of non-equilibrium iptv data set synthetic method it is characterised in that: specifically include following step
Rapid:
Step 1: find out minority class sample set xminorIn each sample point xiCorresponding k neighbour set si, wherein k is natural number, i
=1 ... n, xi∈xminor；K neighbour collection is combined into apart from xiThe set that k nearest sample is formed；
Step 2, the characteristic of k each the minority class sample of neighbour's set analysis being obtained according to step 1, and then it is classified as noise
Collection, safe collection and dangerous collection three classes；
Step 3, the sample that noise is concentrated does not process, and calculates the sample size in safe collection and the dangerous sample size concentrated
Between ratio t；
Step 4, produces an equally distributed random number b obeying on interval [0,1]；If b is ∈ [0, t/ (t+1)], then select
The dangerous all samples concentrated generate new minority class sample as input, the smote algorithm sending into standard；Conversely, then selecting
All samples in safe collection generate new minority class sample as input, the smote algorithm sending into standard；
Step 5, original minority class sample and newly-generated minority class sample are combined the new minority class set of composition.
2. the synthetic method of the minority class sample on a kind of non-equilibrium iptv data set according to claim 1, its feature
It is: described step 2 specifically comprises the steps of:
Step 2.1, counts siIn belong to many several classes ofs xmajorNumber of samples, use | si∩xmajor| to represent, it represents most
Class sample set xmajorAnd siCommon factor in number of samples；
Step 2.2, judges | si∩xmajor| residing interval, it is specifically divided into three kinds of situations:
If | si∩xmajor|=k, then current sample xiIt is in most apoplexy due to endogenous wind, it is believed that it is noise for classification problem；
xminorIn all samples composition safe collections meeting this condition；
If 0≤| si∩xmajor| < 0.5k then shows current sample xiDangerous very little by misclassification；xminorIn all meet this
The sample composition safe collection of condition；
If 0.5k≤| si∩xmajor| < k then shows current sample xiExist by the danger of misclassification；xminorIn all meet this
The dangerous collection of sample composition of condition.
3. the synthetic method of the minority class sample on a kind of non-equilibrium iptv data set according to claim 1, its feature
It is: in step 4, the concrete calculating process of algorithm of described smote is as follows: sets current sample as xi, near from the k of this sample
Adjacent set siOne sample x of middle random selectionj, produce an equally distributed random number δ of obedience from interval [0,1], then newly-generated
Minority class sample be: xnew=xi+δ×(xj-xi).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610753263.9A CN106372655A (en) | 2016-08-26 | 2016-08-26 | Synthetic method for minority class samples in non-balanced IPTV data set |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610753263.9A CN106372655A (en) | 2016-08-26 | 2016-08-26 | Synthetic method for minority class samples in non-balanced IPTV data set |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106372655A true CN106372655A (en) | 2017-02-01 |
Family
ID=57903855
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610753263.9A Pending CN106372655A (en) | 2016-08-26 | 2016-08-26 | Synthetic method for minority class samples in non-balanced IPTV data set |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106372655A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109871862A (en) * | 2018-12-28 | 2019-06-11 | 北京航天测控技术有限公司 | A kind of failure prediction method based on synthesis minority class over-sampling and deep learning |
| CN111782904A (en) * | 2019-12-10 | 2020-10-16 | 国网天津市电力公司电力科学研究院 | Improved SMOTE algorithm-based unbalanced data set processing method and system |
| CN112365060A (en) * | 2020-11-13 | 2021-02-12 | 广东电力信息科技有限公司 | Preprocessing method for power grid internet of things perception data |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102945280A (en) * | 2012-11-15 | 2013-02-27 | 翟云 | Unbalanced data distribution-based multi-heterogeneous base classifier fusion classification method |
| CN103500159A (en) * | 2013-09-06 | 2014-01-08 | 西安交通大学 | Method for recognizing topics of nonequilibrium interactive texts based on example obtaining |
| CN103927874A (en) * | 2014-04-29 | 2014-07-16 | 东南大学 | Automatic incident detection method based on under-sampling and used for unbalanced data set |
| CN105589806A (en) * | 2015-12-17 | 2016-05-18 | 北京航空航天大学 | SMOTE+Boosting algorithm based software defect tendency prediction method |
| CN105760889A (en) * | 2016-03-01 | 2016-07-13 | 中国科学技术大学 | Efficient imbalanced data set classification method |
-
2016
- 2016-08-26 CN CN201610753263.9A patent/CN106372655A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102945280A (en) * | 2012-11-15 | 2013-02-27 | 翟云 | Unbalanced data distribution-based multi-heterogeneous base classifier fusion classification method |
| CN103500159A (en) * | 2013-09-06 | 2014-01-08 | 西安交通大学 | Method for recognizing topics of nonequilibrium interactive texts based on example obtaining |
| CN103927874A (en) * | 2014-04-29 | 2014-07-16 | 东南大学 | Automatic incident detection method based on under-sampling and used for unbalanced data set |
| CN105589806A (en) * | 2015-12-17 | 2016-05-18 | 北京航空航天大学 | SMOTE+Boosting algorithm based software defect tendency prediction method |
| CN105760889A (en) * | 2016-03-01 | 2016-07-13 | 中国科学技术大学 | Efficient imbalanced data set classification method |
Non-Patent Citations (2)
| Title |
|---|
| HUI HAN等: ""Borderline-SMOTE:A new over-sampling method in imbalanced data sets learning"", 《ADVANCES IN INTELLIGENT COMPUTING》 * |
| 王璐林: ""面向不平衡样本的Boosting分类算法研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109871862A (en) * | 2018-12-28 | 2019-06-11 | 北京航天测控技术有限公司 | A kind of failure prediction method based on synthesis minority class over-sampling and deep learning |
| CN111782904A (en) * | 2019-12-10 | 2020-10-16 | 国网天津市电力公司电力科学研究院 | Improved SMOTE algorithm-based unbalanced data set processing method and system |
| CN111782904B (en) * | 2019-12-10 | 2023-10-27 | 国网天津市电力公司电力科学研究院 | Unbalanced data set processing method and system based on improved SMOTE algorithm |
| CN112365060A (en) * | 2020-11-13 | 2021-02-12 | 广东电力信息科技有限公司 | Preprocessing method for power grid internet of things perception data |
| CN112365060B (en) * | 2020-11-13 | 2024-01-26 | 广东电力信息科技有限公司 | Preprocessing method for network Internet of things sensing data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111817982B (en) | Encrypted flow identification method for category imbalance | |
| CN107330477A (en) | A kind of improvement SMOTE resampling methods classified for lack of balance data | |
| CN108229550B (en) | Cloud picture classification method based on multi-granularity cascade forest network | |
| CN105760889A (en) | Efficient imbalanced data set classification method | |
| CN102420723A (en) | Anomaly detection method for various kinds of intrusion | |
| CN107038167A (en) | Big data excavating analysis system and its analysis method based on model evaluation | |
| CN106228389A (en) | Network potential usage mining method and system based on random forests algorithm | |
| CN110012029A (en) | A kind of method and system for distinguishing encryption and non-encrypted compression flow | |
| CN106372655A (en) | Synthetic method for minority class samples in non-balanced IPTV data set | |
| CN107273387A (en) | Towards higher-dimension and unbalanced data classify it is integrated | |
| CN108243435B (en) | Parameter optimization method and device in LTE cell scene division | |
| CN109981474A (en) | A kind of network flow fine grit classification system and method for application-oriented software | |
| CN116582133B (en) | Intelligent management system for data in transformer production process | |
| CN114239807A (en) | RFE-DAGMM-based high-dimensional data anomaly detection method | |
| CN104484412A (en) | Big data analysis system based on multiform processing | |
| CN103780588A (en) | User abnormal behavior detection method in digital home network | |
| CN111145027A (en) | Suspected money laundering transaction identification method and device | |
| CN106251241A (en) | A kind of feature based selects the LR Bagging algorithm improved | |
| CN109376752A (en) | A kind of PTM-WKNN classification method and device based on unbalanced dataset | |
| CN109213865A (en) | A kind of software bug report categorizing system and classification method | |
| CN106056160B (en) | User fault reporting prediction method under unbalanced IPTV data set | |
| CN111600878A (en) | Low-rate denial of service attack detection method based on MAF-ADM | |
| CN107770813B (en) | LTE uplink interference classification method based on PCA and two-dimensional skewness characteristics | |
| CN110097120A (en) | Network flow data classification method, equipment and computer storage medium | |
| CN108494620B (en) | Network service flow characteristic selection and classification method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170201 |
|
| RJ01 | Rejection of invention patent application after publication |