CN108494620A - Network service flow feature selecting and sorting technique based on multiple target Adaptive evolvement arithmetic - Google Patents
Network service flow feature selecting and sorting technique based on multiple target Adaptive evolvement arithmetic Download PDFInfo
- Publication number
- CN108494620A CN108494620A CN201810169202.7A CN201810169202A CN108494620A CN 108494620 A CN108494620 A CN 108494620A CN 201810169202 A CN201810169202 A CN 201810169202A CN 108494620 A CN108494620 A CN 108494620A
- Authority
- CN
- China
- Prior art keywords
- feature
- classification
- evolvement
- arithmetic
- feature selecting
- 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.)
- Granted
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L43/00—Arrangements for monitoring or testing data switching networks
- H04L43/06—Generation of reports
- H04L43/062—Generation of reports related to network traffic
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L43/00—Arrangements for monitoring or testing data switching networks
- H04L43/02—Capturing of monitoring data
- H04L43/026—Capturing of monitoring data using flow identification
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L43/00—Arrangements for monitoring or testing data switching networks
- H04L43/02—Capturing of monitoring data
- H04L43/028—Capturing of monitoring data by filtering
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L43/00—Arrangements for monitoring or testing data switching networks
- H04L43/08—Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters
- H04L43/0876—Network utilisation, e.g. volume of load or congestion level
- H04L43/0894—Packet rate
Abstract
The invention discloses a kind of network service flow feature selectings and sorting technique based on multiple target Adaptive evolvement arithmetic, this method is ranked up feature first with information gain-ratio, filter out part extraneous features, achieve the purpose that quick dimensionality reduction, then feature space is scanned for Adaptive evolvement arithmetic, the feature for being used in combination information gain-ratio in the top chooses optimal feature subset using two object functions of inconsistent rate and character subset dimension as evaluation function as initial population.Adaptive crossover and mutation maintains population diversity, and ensure that the convergence capabilities of algorithm.The present invention utilizes three layers of KNN sorter models of design to online SD live video, web page browsing simultaneously(Baidu), online audio, web page browsing(sina), voice-over-net chat, online SD six kinds of multimedia service streams of non-live video classify.The experimental results showed that this method has higher classification accuracy than existing method.
Description
Technical field
The invention belongs to pattern recognition and classification technical fields, and in particular to one kind being based on multiple target Adaptive evolvement arithmetic
Network service flow feature selecting and sorting technique.
Background technology
In recent years, With the fast development of internet, the network flow classification of precise and high efficiency is the important foundation of network management.
The diversity of network multimedia traffic flow types brings huge challenge to its classification and identification.Traditional stream sorting technique master
It to include three kinds:Method based on port, deep packet inspection method and the method based on media stream statistical nature.But with
Data encryption, the appearance of new application and the use of dynamic port, first two sorting technique will be no longer applicable in.Nowadays, most of
Researcher pays close attention to including machine learning classifications methods such as decision tree, SVM (SupportVectorMachine) and C5.0.
In practical applications, intrinsic dimensionality is often very high, and uncorrelated and redundancy feature presence is easy to cause model training
Required time is long and complexity is higher, is not easy to promote.Feature selecting can filter out uncorrelated and redundancy feature, quick to realize
Dimensionality reduction improves model accuracy.Feature selecting algorithm can be divided into filter-type (Filter), encapsulation according to the difference of evaluation function
Type (Wrapper) and embedded type (Embed).The process of filter-type feature selecting be it is independent, it is unrelated with specific grader.
Encapsulation type is to be combined together the design of feature selecting and grader, goes to assess selected feature with classification accuracy, to
Select optimal subset.And embedded type is the mould as obtained by analysis using feature selection approach as a part for classifier training
The classification results of type choose subset.Feature selection approach common at present has information gain-ratio (GR), Pearson correlation coefficients, card
Fang Tongji etc..It when intrinsic dimensionality is excessively high, needs to improve efficiency by searching algorithm, has many searching algorithm applications in recent years
Before feature selecting, such as sequence R selection algorithms etc. are removed to selection (SFS), sequence backward selection (SBS) and increasing L.Intelligence at present
Optimizing Search algorithm has become the hot spot of research, and such as evolution algorithmic (EA), particle cluster algorithm etc. obtains in terms of feature selecting
It is widely applied.But these methods all only considered single criterion in search characteristics subset, there is no consider selected feature
The radix of subset, they belong to single goal feature selection approach.
Multiple-objection optimization can be from the quality of multiple angle evaluating characteristic subsets, and using these evaluation indexes as target letter
Number is carried out at the same time optimization.It is inspired by nature biotechnology evolutionary process, researcher proposes for solving multi-objective optimization question
Multi-objective evolutionary algorithm, such as non-dominant radial base evolution algorithmic (ENORA).However, when intrinsic dimensionality is higher, it is uncorrelated
The time complexity of multiple-objection optimization can be increased with redundancy feature.For evolution algorithmic, initialization, intersection and the variation of population are generally
Rate selects the improper convergence capabilities that can all reduce final classification accuracy and algorithm.And most multiple target feature selectings are calculated at present
One object function of method is the accuracy rate of grader, so convergence rate is slower, run time is longer.
Invention content
The technical problem to be solved by the present invention is to:
The shortcomings that overcome algorithm above, the present invention propose a kind of network service based on multiple target Adaptive evolvement arithmetic
Flow feature selecting and sorting technique.
The present invention uses following technical scheme to solve above-mentioned technical problem:
The present invention proposes a kind of network service flow feature selecting and sorting technique based on multiple target Adaptive evolvement arithmetic,
It comprises the steps of:
(1) data collection and pretreatment:The data flow sample of various multimedia services on internet is acquired, is then carried out pre-
Processing operation;
(2) feature selecting and analysis:The statistical nature of above-mentioned network data flow sample is analyzed, selects effectively to distinguish
The feature of Business Stream combines;
(3) service stream classification and inspection:Classification experiments are carried out to network multimedia Business Stream using three layers of KNN graders,
It obtains classification results, and calculates whole classification accuracy rate.
Further, the network multimedia Business Stream feature selecting proposed by the present invention based on multiple target Adaptive evolvement arithmetic
With sorting technique, the data collection is specifically included with pretreatment operation:
(2.1) in open internet environment, pass through more matchmakers needed for the WireShark crawls of network package analysis software
Initial data, is then converted into the five-tuple text formatting of standard by body traffic data, and the five-tuple text formatting includes
The time of data packet arrival, source IP address, purpose IP address, agreement, data packet packet size;
(2.2) it carries out basic statistical nature to the standard quintuple file of initial multimedia Business Stream to calculate, the system
Counting feature includes:Uplink/downlink packet size, uplink/downlink packet size information entropy, whole packet size, between uplink/downlink packet arrival time
Every, the ratio between downlink data packet rate, downlink byte-rate and uplink and downlink byte number.
Further, the network multimedia Business Stream feature selecting proposed by the present invention based on multiple target Adaptive evolvement arithmetic
With sorting technique, the feature selecting is specifically included with analysis:
(3.1) all features are ranked up using information gain-ratio, filter out the feature less than relevance threshold;
(3.2) coding selection:The binary coding that length is characterized quantity N is chosen, each coding individual is by a string of bits
Position composition;Any bit is worth all there are two value and represents selection this feature for 1, is worth to represent for 0 and not select;Each individual is expressed as:WhereincIAnd mIRespectively represent each coding
The discrete parameter of adaptive crossover and mutation is executed in individual;
(3.3) initialization of population:The empty population P of initialization0, when number of individuals is less than Population Size popsize in population, follow
Ring executes the value of the random initializtion q in [1, N] range, and individual chooses information gain-ratio q feature in the top, i.e., will be right
It to the positions N is 0 that the preceding positions q answered, which are 1, q+1, which is added population P0;
(3.4) there are two fitness function f by each individual I1(I) and f2(I), correspond to two targets of multiple-objection optimization
Function;Wherein f1(I) it is inconsistent rate, f2(I) selected Characteristic Number is represented;
(3.5) parent is selected:Parent is selected according to the crowding distance of individual;
(3.6) adaptive to intersect:
Fixed crossover probability pc, for any two individuals I and J in t generations, if the Bernoulli random variable is with pcProbability take
1, then by cJIt is random to be set to 0 or 1, and by cJValue be assigned to cI;If cJValue be 0, then do not intersect, if it is 1 execute uniformly hand over
Fork;
The new individual generated will be intersected, Q in auxiliary population is addedtIn;
(3.7) TSP question:
Fixed mutation probability pm, for t generation individual I, if the Bernoulli random variable is with pmProbability take 1, then by mIAt random
It is set to 0 or 1;If mIValue be 0, then without variation, if it is 1 carry out single-point overturning variation;
Q is added in the new individual that variation is generatedtFor in population, and by parent PtAnd QtMerge into auxiliary population Rt;
To population RtIn all individuals be ranked up according to the grade and crowding distance of object function, before selection
Next generation P is arrived in popsize individual survivalt+1;
Execute t=t+1;
(3.8) if meeting maximum iteration gen or inconsistent rate remains unchanged in an iterative process, optimal spy is exported
Levy subset;Otherwise step (3.4) is repeated to step (3.7).
Further, the network multimedia Business Stream feature selecting proposed by the present invention based on multiple target Adaptive evolvement arithmetic
With sorting technique, the inconsistent rate refers to:A feature group in sample instance is collectively referred to as a pattern, character subset
The inconsistent number of all patterns subtracts the sample of the most a certain class label of occurrence number for the total sample number of pattern appearance
Number, inconsistent rate are equal to inconsistent number divided by total sample number.
Further, the network multimedia Business Stream feature selecting proposed by the present invention based on multiple target Adaptive evolvement arithmetic
With sorting technique, the relevance threshold in step (3.1) is 0.4, and the corresponding three-layer classification devices of N in step (3.2) are followed successively by
25,26,13, the crossover probability p in step (3.6)cWith the mutation probability p in step (3.7)mIt is 0.1, in step (3.7)
Popsize=100, the maximum iteration gen in step (3.8) are 10.
Further, the network multimedia Business Stream feature selecting proposed by the present invention based on multiple target Adaptive evolvement arithmetic
With sorting technique, service stream classification step specifically includes:
(5.1) it uses feature selection approach to carry out feature selecting to initial multimedia Business Stream, and carries out KNN points of first layer
Media stream is divided into 4 classes by class:C1, C2, C3, C4;Wherein C1 is online audio, and C2 is Online Video, and C3 is web page browsing,
C4 chats for voice-over-net;
(5.2) video stream characteristics of the C2 to classify to last layer reuse feature selection approach and carry out feature choosing
It selects, and carries out second layer KNN classification, obtain classification results C21, C22;
(5.3) feature selection approach is reused to the data flow characteristics of step (5.1) classification results C3 and carries out feature choosing
It selects, and carries out second KNN classification of the second layer, obtain classification results C31, C32;
(5.4) statistical classification output is as a result, calculate whole classification accuracy rate.
Further, the network multimedia Business Stream feature selecting proposed by the present invention based on multiple target Adaptive evolvement arithmetic
With sorting technique, the classification results C21 is online live video, and C22 is online non-live video;C31 web page contents are text
Word and picture, C32 web page contents are word, picture and video.
The present invention has the following technical effects using above technical scheme is compared with the prior art:
1, the present invention uses the characteristics of the multimedia selection method of multiple target Adaptive evolvement arithmetic relative to single goal feature
Selection algorithm not only considers classification accuracy and considers selected Characteristic Number;It is calculated compared to existing multiple target feature selecting
Method has lower computation complexity, and convergence rate is very fast, can effectively reduce the time in feature selection process and space
Expense improves the efficiency of feature selecting.
2, the method that the present invention uses multistratum classification to multimedia service devises a kind of three layers of KNN cascade classifiers, first
Validity feature combination is chosen first with the feature selection approach of the present invention, then three-layer classification device using the present invention is divided
Class.Compared to existing multilayer svm classifier method, there is preferable classification accuracy rate.
3, the present invention is to online SD live video, web page browsing (Baidu), online audio, web page browsing (sina), net
Six kinds of network voice-enabled chat, the non-live video of online SD multimedia service streams carry out feature selecting, then three layers of KNN are utilized to classify
Device is classified.The experimental results showed that this method has higher identification relative to tri- kinds of feature selecting algorithms of GR, EA, ENORA
Rate, overall accuracy are 98.6%.
Description of the drawings
Fig. 1 is the flow diagram of sorting technique of the present invention.
Fig. 2 is effective proof diagram of the feature combination selected by the present invention, wherein (a) is four kinds of network service flows under
Two-dimensional distribution in row packet size maximum value and uplink byte number is (b) two kinds of video types in downstream packets size maximum value
It is (c) two kinds of web page browsing types in downstream packets size maximum value and downlink byte with the two-dimensional distribution on uplink byte number
Two-dimensional distribution in the rate degree of bias.
Fig. 3 is that the present invention, GR, EA and ENORA pairs of six kinds of multimedia service classification accuracy rates compare figure.
Specific implementation mode
Technical scheme of the present invention is described in further detail below in conjunction with the accompanying drawings:
Those skilled in the art of the present technique are it is understood that unless otherwise defined, all terms used herein (including skill
Art term and scientific terminology) there is meaning identical with the general understanding of the those of ordinary skill in fields of the present invention.Also
It should be understood that those terms such as defined in the general dictionary should be understood that with in the context of the prior art
The consistent meaning of meaning, and unless defined as here, will not be explained with the meaning of idealization or too formal.
As shown in Figure 1, the present invention proposes a kind of multimedia service stream feature based on multiple target Adaptive evolvement arithmetic
Selection and sorting technique, this method include data acquisition and the pretreatment of multimedia service stream, are based on multiple target Adaptive evolution
The multimedia service stream feature selecting of algorithm, three layers of KNN cascade sorts output statistical result etc., include the following steps:
Step 1:Data collection and pretreatment, the specific steps are:
(1) in open internet environment, pass through the multimedia needed for the WireShark crawls of network package analysis software
Initial data, is then converted into the five-tuple text formatting of standard by traffic data, i.e., data packet reaches time, source IP
Location, purpose IP address, agreement, data packet packet size;
(2) it carries out basic statistical nature to the standard quintuple file of initial multimedia Business Stream to calculate, these features
Including:Uplink/downlink packet size, uplink/downlink packet size information entropy, whole packet size, uplink/downlink packet arrival time time interval,
The ratio between downlink data packet rate, downlink byte-rate and uplink and downlink byte number.
Step 2:Feature selecting and analysis, the specific steps are:
(1) all features are ranked up using information gain-ratio, filter out the feature less than relevance threshold 0.4;
(2) coding selection:Binary coding is chosen, each individual is made of (length is characterized quantity N) a string of bits.
Any bit is worth all there are two value and represents selection this feature for " 1 ", is worth to represent for 0 and not select;
(3) initialization of population:The value of random initializtion q in [1, N] range chooses information gain-ratio q in the top
A feature is as initialization population P0, i.e., by the corresponding preceding positions q be 1, q+1 to the positions N be 0;
(4) there are two fitness function f by each individual I1(I) and f2(I), correspond to two target letters of multiple-objection optimization
Number.Wherein f1(I) it is inconsistent rate, a feature group in sample instance is collectively referred to as a pattern, character subset owns
The inconsistent number of pattern, the total sample number for being equal to pattern appearance subtract the sample of the most a certain class label of occurrence number
Number, inconsistent rate are equal to inconsistent number divided by total sample number;f2(I) selected Characteristic Number is represented;
(5) parent is selected:Parent is selected according to the crowding distance of individual;
(6) adaptive to intersect:Crossover probability p fixed firstc=0.1, then for PtAny two individuals I and J in generation, if
The Bernoulli random variable is with pcProbability take 1, then by cJIt is random to be set to 0 or 1, and by cJValue be assigned to cI.If cJValue be 0,
Do not intersect then, uniform crossover is executed if it is 1.The new individual generated will be intersected, P is addedt+1For in population;
(7) TSP question:Mutation probability p fixed firstm=0.1, for PtGeneration individual I, if the Bernoulli random variable
With pmProbability take 1, then by mIIf being set to 0 or 1. m at randomIValue be 0, then without variation, carry out single-point if it is 1
Overturning variation.P is added in the new individual that variation is generatedt+1For in population, t=t+1 is executed;
(8) it if meeting maximum iteration gen=10 or inconsistent rate remains unchanged in an iterative process, exports optimal
Character subset;Otherwise step (4) is repeated to step (7).
We devise three layers of KNN cascade classifier models in an experiment, this model can be in the grader of every level-one
The feature combination selected using the method for the present invention identifies certain certain types of applied business.The KNN grader masters of first layer
To be used for identifying online audio (QQ Music), Online Video (live streaming and non-live streaming), web page browsing, voice-over-net chat
(Skype), best feature is combined as downstream packets size maximum value, uplink byte number.In order to facilitate observation, we are to Fig. 2's
(a) it has done and has taken log operations.Belong to interactive audio from Skype known to (a) of Fig. 2, so uplink byte number is clear higher than webpage
It lookes at QQ Music and less than online live video, it can be with efficient identification using downstream packets size maximum value and uplink byte number
Skype and QQ Music.
The video that obtains of first layer classification is further divided into online live video and online non-by the KNN graders of the second layer
Live video.Best feature combination:Uplink byte number.CBox belongs to live video type, and client and server interacts number
According to significantly more than non-live broadcast service youku videos, so it can be seen that can be incited somebody to action by feature uplink byte number from (b) of Fig. 2
Live streaming and non-live streaming video traffic are kept completely separate.
The web page browsing that obtains of first layer classification is further divided into Baidu by the KNN graders of third layer, and (web page contents are
Word and picture) and sina (web page contents are word, picture, video).Best feature combination:Downstream packets size maximum value and
The downlink byte-rate degree of bias.Since video class business data packet is more than other Business Streams, and sina browsing contents include video class,
So the downstream packets size maximum value of sina is slightly larger than Baidu Business Streams.As shown in (c) of Fig. 2, feature downstream packets size is maximum
Value and the downlink byte-rate degree of bias can accurately identify sina and Baidu Business Streams
Step 3, service stream classification and inspection, the specific steps are:
(1) it uses feature selection approach to carry out feature selecting to initial multimedia Business Stream, and carries out KNN points of first layer
Media stream is divided into 4 class C1, C2, C3, C4 by class;Wherein C1 is online audio (QQ Music), and C2 is Online Video (live streaming
With non-live streaming), C3 is web page browsing, and C4 is that voice-over-net chats (Skype);
(2) video stream characteristics of the C2 to classify to last layer reuse feature selection approach and carry out feature selecting,
And second layer KNN classification is carried out, obtain classification results C21, C22;Wherein, C21 is online live video, and C22 is online non-straight
Broadcast video;
(3) feature selection approach is reused to the data flow characteristics of step (1) classification results C3 and carries out feature selecting, and
The second KNN classification for carrying out the second layer, obtains classification results C31, C32;Wherein, C31 is that (web page contents are word to Baidu
And picture), C32 is sina (web page contents are word, picture, video);
(4) statistical classification output is as a result, calculate whole classification accuracy rate.
In an experiment we use eighty percent discount cross validation, and by the present invention classification results same GR, EA, ENORA result
It is compared.As can be seen from Figure 3, the method for the present invention has a highest whole classification accuracy rate, and up to 98.6%.
The above is only some embodiments of the present invention, it is noted that for the ordinary skill people of the art
For member, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also answered
It is considered as protection scope of the present invention.
Claims (7)
1. network service flow feature selecting and sorting technique based on multiple target Adaptive evolvement arithmetic, which is characterized in that include
Following steps:
(1) data collection and pretreatment:The data flow sample for acquiring various multimedia services on internet, is then pre-processed
Operation;
(2) feature selecting and analysis:The statistical nature of above-mentioned network data flow sample is analyzed, effective differentiated service is selected
The feature of stream combines;
(3) service stream classification and inspection:Classification experiments are carried out to network multimedia Business Stream using three layers of KNN graders, are obtained
Classification results, and calculate whole classification accuracy rate.
2. the network multimedia Business Stream feature selecting according to claim 1 based on multiple target Adaptive evolvement arithmetic with
Sorting technique, which is characterized in that the data collection is specifically included with pretreatment operation:
(2.1) in open internet environment, pass through the multimedia industry needed for the WireShark crawls of network package analysis software
Initial data, is then converted into the five-tuple text formatting of standard, the five-tuple text formatting includes data by business flow data
Wrap the time reached, source IP address, purpose IP address, agreement, data packet packet size;
(2.2) it carries out basic statistical nature to the standard quintuple file of initial multimedia Business Stream to calculate, the statistics is special
Sign includes:Uplink/downlink packet size, uplink/downlink packet size information entropy, whole packet size, uplink/downlink packet interarrival times, under
The ratio between row packet rate, downlink byte-rate and uplink and downlink byte number.
3. the network multimedia Business Stream feature selecting according to claim 1 based on multiple target Adaptive evolvement arithmetic with
Sorting technique, which is characterized in that the feature selecting is specifically included with analysis:
(3.1) all features are ranked up using information gain-ratio, filter out the feature less than relevance threshold;
(3.2) coding selection:The binary coding that length is characterized quantity N is chosen, each coding individual is by a string of bit hytes
At;Any bit is worth all there are two value and represents selection this feature for 1, is worth to represent for 0 and not select;Each individual is expressed as:WhereinI=1 ..., N, cI∈{0,1},mI∈{0,1};cIAnd mIIt respectively represents every
The discrete parameter of adaptive crossover and mutation is executed in a coding individual;
(3.3) initialization of population:The empty population P of initialization0, when number of individuals is less than Population Size popsize in population, cycle is held
The value of row random initializtion q in [1, N] range, individual choose information gain-ratio q feature in the top, i.e., will be corresponding
It to the positions N is 0 that the preceding positions q, which are 1, q+1, which is added population P0;
(3.4) there are two fitness function f by each individual I1(I) and f2(I), correspond to two object functions of multiple-objection optimization;
Wherein f1(I) it is inconsistent rate, f2(I) selected Characteristic Number is represented;
(3.5) parent is selected:Parent is selected according to the crowding distance of individual;
(3.6) adaptive to intersect:
Fixed crossover probability pc, for any two individuals I and J in t generations, if the Bernoulli random variable is with pcProbability take 1, then
By cJIt is random to be set to 0 or 1, and by cJValue be assigned to cI;If cJValue be 0, then do not intersect, uniform crossover executed if it is 1;
The new individual generated will be intersected, Q in auxiliary population is addedtIn;
(3.7) TSP question:
Fixed mutation probability pm, for t generation individual I, if the Bernoulli random variable is with pmProbability take 1, then by mIIt is set at random
0 or 1;If mIValue be 0, then without variation, if it is 1 carry out single-point overturning variation;
Q is added in the new individual that variation is generatedtFor in population, and by parent PtAnd QtMerge into auxiliary population Rt;
To population RtIn all individuals be ranked up according to the grade and crowding distance of object function, popsize before choosing
Body is survived to next-generation Pt+1;
Execute t=t+1;
(3.8) if meeting maximum iteration gen or inconsistent rate remains unchanged in an iterative process, optimal characteristics is exported
Collection;Otherwise step (3.4) is repeated to step (3.7).
4. the network multimedia Business Stream feature selecting according to claim 3 based on multiple target Adaptive evolvement arithmetic with
Sorting technique, which is characterized in that the inconsistent rate refers to:A feature group in sample instance is collectively referred to as a pattern,
The inconsistent number of all patterns of character subset, the total sample number occurred for the pattern subtract the most a certain category of occurrence number
The sample number of label, inconsistent rate are equal to inconsistent number divided by total sample number.
5. the network multimedia Business Stream feature selecting according to claim 3 based on multiple target Adaptive evolvement arithmetic with
Sorting technique, which is characterized in that the relevance threshold in step (3.1) is 0.4, the corresponding three-layer classifications of N in step (3.2)
Device is followed successively by 25,26,13, the crossover probability p in step (3.6)cWith the mutation probability p in step (3.7)mIt is 0.1, step
(3.7) popsize=100 in, the maximum iteration gen in step (3.8) are 10.
6. the network multimedia Business Stream feature selecting according to claim 1 based on multiple target Adaptive evolvement arithmetic with
Sorting technique, which is characterized in that service stream classification step specifically includes:
(5.1) it uses feature selection approach to carry out feature selecting to initial multimedia Business Stream, and carries out first layer KNN classification,
Media stream is divided into 4 classes:C1, C2, C3, C4;Wherein C1 is online audio, and C2 is Online Video, and C3 is web page browsing, and C4 is
Voice-over-net is chatted;
(5.2) video stream characteristics of the C2 to classify to last layer reuse feature selection approach and carry out feature selecting, and
Second layer KNN classification is carried out, classification results C21, C22 are obtained;
(5.3) feature selection approach is reused to the data flow characteristics of step (5.1) classification results C3 and carries out feature selecting, and
The second KNN classification for carrying out the second layer, obtains classification results C31, C32;
(5.4) statistical classification output is as a result, calculate whole classification accuracy rate.
7. the network multimedia Business Stream feature selecting according to claim 6 based on multiple target Adaptive evolvement arithmetic with
Sorting technique, which is characterized in that the classification results C21 is online live video, and C22 is online non-live video;C31 webpages
Content is word and picture, and C32 web page contents are word, picture and video.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810169202.7A CN108494620B (en) | 2018-02-28 | 2018-02-28 | Network service flow characteristic selection and classification method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810169202.7A CN108494620B (en) | 2018-02-28 | 2018-02-28 | Network service flow characteristic selection and classification method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108494620A true CN108494620A (en) | 2018-09-04 |
CN108494620B CN108494620B (en) | 2021-07-27 |
Family
ID=63341158
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810169202.7A Active CN108494620B (en) | 2018-02-28 | 2018-02-28 | Network service flow characteristic selection and classification method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108494620B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112580606A (en) * | 2020-12-31 | 2021-03-30 | 安徽大学 | Large-scale human body behavior identification method based on clustering grouping |
CN113079427A (en) * | 2021-04-28 | 2021-07-06 | 北京航空航天大学 | ASON network service availability evaluation method based on network evolution model |
CN115049019A (en) * | 2022-07-25 | 2022-09-13 | 湖南工商大学 | Method and device for evaluating arsenic adsorption performance of metal organic framework and related equipment |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120054184A1 (en) * | 2010-08-24 | 2012-03-01 | Board Of Regents, The University Of Texas System | Systems and Methods for Detecting a Novel Data Class |
US20130108131A1 (en) * | 2007-05-29 | 2013-05-02 | University Of Iowa Research Foundation | Methods And Systems For Determining Optimal Features For Classifying Patterns Or Objects In Images |
CN105260746A (en) * | 2015-10-09 | 2016-01-20 | 乔善平 | Expandable multilayer integrated multi-mark learning system |
CN105787437A (en) * | 2016-02-03 | 2016-07-20 | 东南大学 | Vehicle brand type identification method based on cascading integrated classifier |
CN106897733A (en) * | 2017-01-16 | 2017-06-27 | 南京邮电大学 | Video stream characteristics selection and sorting technique based on particle swarm optimization algorithm |
-
2018
- 2018-02-28 CN CN201810169202.7A patent/CN108494620B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130108131A1 (en) * | 2007-05-29 | 2013-05-02 | University Of Iowa Research Foundation | Methods And Systems For Determining Optimal Features For Classifying Patterns Or Objects In Images |
US20120054184A1 (en) * | 2010-08-24 | 2012-03-01 | Board Of Regents, The University Of Texas System | Systems and Methods for Detecting a Novel Data Class |
CN105260746A (en) * | 2015-10-09 | 2016-01-20 | 乔善平 | Expandable multilayer integrated multi-mark learning system |
CN105787437A (en) * | 2016-02-03 | 2016-07-20 | 东南大学 | Vehicle brand type identification method based on cascading integrated classifier |
CN106897733A (en) * | 2017-01-16 | 2017-06-27 | 南京邮电大学 | Video stream characteristics selection and sorting technique based on particle swarm optimization algorithm |
Non-Patent Citations (2)
Title |
---|
CHANGJING SHANG,: ""Combining support vector machines and information gain ranking for classification of mars mcmurdo panorama images"", 《PROCEEDINGS OF 2010 IEEE 17TH INTERNATIONAL CONFERENCE ON IMAGE PROCESSING》 * |
冯茂,: ""基于粒子群优化算法的视频流特征选择方法"", 《南京邮电大学学报》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112580606A (en) * | 2020-12-31 | 2021-03-30 | 安徽大学 | Large-scale human body behavior identification method based on clustering grouping |
CN112580606B (en) * | 2020-12-31 | 2022-11-08 | 安徽大学 | Large-scale human body behavior identification method based on clustering grouping |
CN113079427A (en) * | 2021-04-28 | 2021-07-06 | 北京航空航天大学 | ASON network service availability evaluation method based on network evolution model |
CN113079427B (en) * | 2021-04-28 | 2021-11-23 | 北京航空航天大学 | ASON network service availability evaluation method based on network evolution model |
CN115049019A (en) * | 2022-07-25 | 2022-09-13 | 湖南工商大学 | Method and device for evaluating arsenic adsorption performance of metal organic framework and related equipment |
Also Published As
Publication number | Publication date |
---|---|
CN108494620B (en) | 2021-07-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Homayoun et al. | BoTShark: A deep learning approach for botnet traffic detection | |
CN101605126B (en) | Method and system for classifying and recognizing a plurality of protocols | |
CN109450845B (en) | Detection method for generating malicious domain name based on deep neural network algorithm | |
CN109151880B (en) | Mobile application flow identification method based on multilayer classifier | |
Yang et al. | TLS/SSL encrypted traffic classification with autoencoder and convolutional neural network | |
CN111565156B (en) | Method for identifying and classifying network traffic | |
CN105809190A (en) | Characteristic selection based SVM cascade classifier method | |
CN108650194A (en) | Net flow assorted method based on K_means and KNN blending algorithms | |
CN104244035A (en) | Network video flow classification method based on multilayer clustering | |
CN108494620A (en) | Network service flow feature selecting and sorting technique based on multiple target Adaptive evolvement arithmetic | |
CN111224994A (en) | Botnet detection method based on feature selection | |
Soleymanpour et al. | CSCNN: cost-sensitive convolutional neural network for encrypted traffic classification | |
Soleymanpour et al. | An efficient deep learning method for encrypted traffic classification on the web | |
Feng et al. | BotFlowMon: Learning-based, content-agnostic identification of social bot traffic flows | |
CN112788038A (en) | Method for distinguishing DDoS attack and elephant flow based on PCA and random forest | |
Sarraf | Analysis and detection of ddos attacks using machine learning techniques | |
Dixit et al. | Internet traffic detection using naïve bayes and K-Nearest neighbors (KNN) algorithm | |
CN106557983B (en) | Microblog junk user detection method based on fuzzy multi-class SVM | |
Long et al. | A method of machine learning for social bot detection combined with sentiment analysis | |
CN1612135A (en) | Invasion detection (protection) product and firewall product protocol identifying technology | |
CN115459937A (en) | Method for extracting characteristics of encrypted network traffic packet in distributed scene | |
CN113256507B (en) | Attention enhancement method for generating image aiming at binary flow data | |
CN110417786B (en) | P2P flow fine-grained identification method based on depth features | |
CN115134128A (en) | Method for mining and utilizing new type encrypted network flow packet in distributed scene | |
Lee et al. | Generation of network traffic using WGAN-GP and a DFT filter for resolving data imbalance |
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 |