A business type identification method based on PCA and XGBoost fusion

Through the fusion method of PCA and XGBoost, the problem that traditional business type identification methods cannot recognize encrypted traffic is solved, and efficient and accurate business type identification is achieved, meeting the management and security needs of modern network environments.

CN114048795BActive Publication Date: 2024-11-08NANJING UNIV OF POSTS & TELECOMM
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202111202293.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-10-15
Publication Date
2024-11-08
Estimated Expiration
2041-10-15

AI Technical Summary

Technical Problem

Traditional business type identification methods cannot effectively identify encrypted traffic, and the recognition accuracy is low, which cannot meet the needs of modern network environments for efficient management and security guarantees.

Method used

The fusion method based on PCA and XGBoost is adopted to extract and classify network traffic data through principal component analysis dimensionality reduction and limit gradient enhancement models, optimize learning rate and regularization parameters, and improve the accuracy and efficiency of business type identification.

Benefits of technology

It realizes efficient identification of encrypted traffic, improves the accuracy of business type identification and the complexity of the model, and ensures efficient and reliable identification results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114048795B_ABST
    Figure CN114048795B_ABST
Patent Text Reader

Abstract

The present invention discloses a business type identification method based on PCA and XGBoost fusion, comprising the following steps: step S1, collecting a network traffic data set with a label, the business types of the network traffic data set include HTTP, NTP, DNS, QQ, WeChat, video and email; step S2, performing data cleaning and feature extraction on the network traffic data set in step S1, so as to obtain a network traffic data set containing multidimensional features; step S3, simplifying the multidimensional features in step S2 into low-dimensional features using the method of principal component analysis feature dimensionality reduction, and obtaining a low-dimensional data set with labels; step S4, obtaining a trained extreme gradient boosting XGBoost classification model; step S5, inputting the network traffic data set to be tested into the extreme gradient boosting XGBoost classification model in step S4, and obtaining a business type classification result. The present invention can not only reduce the complexity of the identification method, but also improve the accuracy of business type identification.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of communication network technology, and in particular to a service type identification method based on PCA and XGBoost fusion. Background Art

[0002] With the continuous development of information technology, the scale of Internet traffic is also increasing year by year, and new network services are emerging one after another. Although this has greatly promoted the progress and development of our society, the popularity of these new services has also attracted a large number of customer resources for telecom operators, but the network contains a variety of encrypted traffic, which has a great impact on the underlying traffic model and upper-layer application model of the network. In order to improve network management, improve network services, and ensure the security of the network environment, effectively identifying the encrypted traffic of various application services and building an "operable and manageable" network has become a key research direction now.

[0003] Traditional traffic type identification methods include port-based traffic identification methods and deep packet inspection (DPI)-based traffic identification methods. Port-based traffic identification is to classify network traffic by the known port number in the TCP / UDP packet header. Initially, this method is very effective and easy to implement for real-time traffic classification. However, nowadays, various network applications do not use well-known ports to avoid detection, and some network applications may use dynamic port numbers when in use. Therefore, port-based traffic classification cannot produce real results and the classification accuracy is not high. The traffic identification method based on deep packet inspection (DPI) is essentially a data packet filtering technology. In addition to supporting the parsing of the packet header of the L2 data link layer, the L3 network layer, and the L4 transport layer, DPI also adds the parsing of the L7 application layer payload, which can identify various application types and their contents. However, since most services currently use various encryption technologies to prohibit the inspection of packet payloads, the classification accuracy of deep packet inspection (DPI) is not very high. The current trend is to use machine learning methods to classify IP traffic.

[0004] In recent years, artificial intelligence technology based on machine learning has achieved remarkable achievements in computer vision, natural language processing, speech recognition, and medical imaging, and its performance in many fields is far superior to traditional solutions. This fully demonstrates the scientificity and effectiveness of machine learning in handling classification tasks. Machine learning and data mining technologies have also gradually been applied and developed in the field of cyberspace security. Therefore, machine learning technology also provides the possibility of solving the encrypted traffic classification problem solved by traditional methods. Traditional business type identification methods cannot identify encrypted traffic and have problems such as low identification accuracy. Summary of the invention

[0005] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide a business type identification method based on the fusion of PCA and XGBoost, which can not only reduce the complexity of the identification method, but also improve the accuracy of business type identification.

[0006] The present invention adopts the following technical solutions to solve the above technical problems:

[0007] A business type identification method based on PCA and XGBoost fusion proposed in the present invention includes the following steps:

[0008] Step S1: Collect a network traffic data set with labels, where the service types of the network traffic data set include HTTP, NTP, DNS, QQ, WeChat, video, and email;

[0009] Step S2, performing data cleaning and feature extraction on the network traffic data set in step S1, thereby obtaining a network traffic data set containing multi-dimensional features;

[0010] Step S3, using the principal component analysis feature dimensionality reduction method to simplify the multi-dimensional features in step S2 into low-dimensional features, and obtain a labeled low-dimensional data set;

[0011] Step S4, selecting key performance indicators with high correlation that affect the measurement of business types from the labeled low-dimensional data set in step S3, the key performance indicators constitute a labeled data set, the data set is divided into a training set and a test set, the training set is input into the extreme gradient boosting XGBoost classification model for training, and the learning rate γ and regularization parameter λ of the XGBoost classification model are tuned by an improved parameter tuning method to obtain the most suitable learning rate γ and regularization parameter λ for the network traffic data set, and the XGBoost classification model after parameter tuning is tested to obtain a trained extreme gradient boosting XGBoost classification model;

[0012] The method for tuning the learning rate γ and the regularization parameter λ is as follows:

[0013] Step S4.1, the extreme gradient boosting XGBoost classification model is:

[0014]

[0015] Among them, T refers to the number of leaf nodes, obj refers to the objective function, G j Refers to the first-order derivative of the objective function in the Taylor expansion of the jth leaf node, H j Refers to the second-order derivative of the Taylor expansion of the objective function at the jth leaf node;

[0016] Step S4.2: Optimize the learning rate γ and regularization parameter λ in step S4.1:

[0017] Step S4.2.1, set the respective search space Φ and search step size μ of the learning rate γ and regularization parameter λ, and set them as follows:

[0018] γ=(γ start ,γ end ,μ γ )

[0019] λ=(λ start ,λ end ,μ λ )

[0020] Among them, γ start and γ end are the search space Φ of the learning rate γ respectively. γ The upper and lower boundaries of μ γ is the search step size of the learning rate γ; λ start and λ end are the search space Θ of the regularization parameter λ respectively. λ Upper and lower boundaries, μ λ is the search step size of the regularization parameter λ;

[0021] Step S4.2.2: Generate a two-dimensional search parameter group matrix H according to the set search space and search step size. S , defined as follows:

[0022]

[0023] Where p is an integer, q is an integer,

[0024] Step S4.2.3: For H in step S4.2.2 S For each parameter group in the XGBoost classification model, the average classification accuracy of each parameter group is evaluated, and the parameter group with the highest average classification accuracy is selected. If there is one parameter group with the highest average classification accuracy, this parameter group is the selected parameter group. If there are multiple parameter groups with the highest average classification accuracy, then select λ from these multiple parameter groups. start +qμ λ The smallest parameter group;

[0025] Step S4.2.4, λ in the parameter group selected in step S4.2.3 start +qμ λ , γ start +pμ γThat is, the optimal learning rate γ and regularization parameter λ of the XGBoost classification model corresponding to the labeled low-dimensional data set in step S3;

[0026] Step S5: input the network traffic data set to be tested into the extreme gradient boosting XGBoost classification model in step S4 to obtain the business type classification result.

[0027] As a further optimization scheme of the business type identification method based on PCA and XGBoost fusion described in the present invention, the multidimensional feature F in step S2 is specifically expressed as follows:

[0028] F=[f1,f2,f3…f d ]

[0029] F contains a vector of d features, f i represents the i-th key characteristic index, d≥i≥1, for f i The maximum values ​​of are normalized respectively, and the specific processing method is as follows:

[0030]

[0031] max(f i ) is the maximum value of the i-th key feature indicator, It is the i-th key feature index after normalization.

[0032] As a further optimization scheme of the business type identification method based on the fusion of PCA and XGBoost described in the present invention, the multi-dimensional features are simplified into low-dimensional features by using the principal component analysis feature dimensionality reduction method in step S3, assuming that the original multi-dimensional features are d-dimensional, and the simplified low-dimensional features are k-dimensional, where k<d;

[0033] Step S3.1, the network traffic data set containing multi-dimensional features in step S2 is formed into a matrix X with d rows and n columns, wherein the network traffic data set containing multi-dimensional features has n samples, and each sample has d-dimensional features;

[0034] Step S3.2, performing a decentralized process on the matrix X of d rows and n columns in step S2 to obtain a decentralized matrix X';

[0035] Step S3.3, calculate the covariance matrix Cov of the matrix X' in step S3.2:

[0036]

[0037] Among them, the superscript T' refers to the transpose;

[0038] Step S3.4, find the eigenvalues ​​and corresponding eigenvectors of the covariance matrix;

[0039] Step S3.5, arrange the obtained eigenvectors into a matrix row by row from top to bottom according to the size of the corresponding eigenvalues, and take the first k rows to form the matrix P;

[0040] Step S3.6, Y=PX, where Y is a low-dimensional feature after being reduced to k dimensions by the principal component analysis feature dimensionality reduction method.

[0041] As a further optimization scheme of the business type identification method based on the fusion of PCA and XGBoost described in the present invention, the actual data set without labels is obtained through the network packet capture tool Wireshark and input into the extreme gradient boosting model XGBoost classification model after PCA dimensionality reduction to obtain the business type classification result.

[0042] As a further optimization scheme of the business type identification method based on the fusion of PCA and XGBoost described in the present invention, in step S4, the key performance indicators with high correlation include source IP, destination IP, source port number, destination port number, protocol type and data length.

[0043] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:

[0044] The method of the present invention solves the problems that traditional business type identification methods cannot identify encrypted traffic and have low identification accuracy. Principal component analysis and XGBoost algorithm are used to balance the accuracy of the test set and the complexity of the algorithm, thereby achieving efficient and reliable business type identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is a flow chart of the business type identification system based on PCA and XGBoost provided by the present invention.

[0046] Figure 2 It is a schematic diagram of feature dimensionality reduction using the principal component analysis (PCA) algorithm. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0048] The present invention proposes to apply the idea of ​​feature dimensionality reduction and deep learning to the business type identification scenario. First, the network traffic big data is obtained through the wireshark network packet capture tool, and a business type identification method based on the fusion of PCA and XGBoost is proposed. Using the idea of ​​feature dimensionality reduction, complex multidimensional features are simplified into low-dimensional features based on the principal component analysis (PCA) algorithm, which reduces the complexity of model training. Then, a large number of reliable labeled data sets are trained using the extreme gradient boosting model (XGBoost), and finally the business type is identified for the actual acquired data traffic. This method combines PCA and XGBoost, which can improve the recognition of business types while reducing the complexity of the algorithm.

[0049] Based on the business type identification application scenario, the present invention proposes a business type identification method based on the fusion of PCA and XGBoost, such as Figure 1 As shown, the method comprises the following steps:

[0050] Step S1: Collect some labeled network traffic data sets through Wireshark, including data sets of HTTP, NTP, DNS, QQ, WeChat, video, email and other business types. The specific representation of data set features is as follows:

[0051] F=[f1,f2,f3…f d ]

[0052] F is a vector containing d features, f i Represents the i-th key feature indicator.

[0053] Step S2: clean and extract features of the labeled data set collected in step 1, remove duplicate samples and invalid samples of incomplete data, and normalize them based on the maximum value of each key feature. The specific processing method is as follows:

[0054]

[0055] in Refers to the normalized i-th key feature index. max(f i ) refers to the maximum value of the i-th key feature indicator.

[0056] Step S3: Before training the classification model, in order to ensure the accuracy of the classification model and reduce the complexity of the data dimension, the principal component analysis feature dimensionality reduction method is used to simplify the complex multidimensional features in step 2 into low-dimensional features, that is, the original d-dimensional data set is converted into a k-dimensional data set, where k < d. In this way, the most important features of the data are retained. The specific processing method of principal component analysis data dimensionality reduction is as follows:

[0057] Step S3.1, the network traffic data set containing multi-dimensional features in step S2 is formed into a matrix X with d rows and n columns, wherein the network traffic data set containing multi-dimensional features has n samples, and each sample has d-dimensional features;

[0058] Step S3.2, performing a decentralized process on the matrix X of d rows and n columns in step S2 to obtain a decentralized matrix X';

[0059] Step S3.3, calculate the covariance matrix Cov of the matrix X' in step S3.2:

[0060]

[0061] Among them, the superscript T' refers to the transpose;

[0062] Step S3.4, find the eigenvalues ​​and corresponding eigenvectors of the covariance matrix;

[0063] Step S3.5, arrange the obtained eigenvectors into a matrix row by row from top to bottom according to the size of the corresponding eigenvalues, and take the first k rows to form the matrix P;

[0064] Step S4: Select key performance indicators with high correlation that affect the measurement of business types from the labeled low-dimensional data set in step S3. The key performance indicators constitute a labeled data set. The data set is divided into a training set and a test set. The training set is input into the extreme gradient boosting model (XGBoost) for training, and the classifier is parameter tuned, and the model after parameter tuning is tested. The specific expression of the extreme gradient boosting model (XGBoost) is as follows:

[0065] Step S4.1, define the objective function:

[0066]

[0067] in is the loss function used to measure the predicted value and the target value y i The difference between them, in this invention, refers to the difference between the predicted business type label and the actual business type label.

[0068] is the regularization term, defined as:

[0069] Where T refers to the number of leaf nodes, λ is the regularization parameter, γ is the learning rate, and w jrepresents the predicted value of the jth leaf node. Since XGBoost is a forward distribution algorithm, the tth result is the result of the previous t-1 times plus the result of the current weak classifier. Therefore, each iteration searches for the CART tree that reduces the loss function the most, so the objective function can be rewritten as:

[0070]

[0071] const means that in the tth round, the regularization term of the first t-1 iterations can be regarded as a constant. The objective function is approximated by Taylor expansion, and we get:

[0072]

[0073] in, Remove the constant term for iteration t

[0074] The objective function is obtained:

[0075]

[0076] Therefore, the objective function only depends on the first-order derivative and second-order derivative of each data on the error function. After processing the regularization term, the final objective function can be rewritten as:

[0077]

[0078] in:

[0079]

[0080]

[0081] I j is defined as the set of indices of samples whose values ​​are associated with leaf node j.

[0082] Assuming that the structure of the decision tree has been determined, the predicted value on each leaf node can be obtained by making the derivative of the loss function zero, which can be written as:

[0083]

[0084] Therefore, the final objective function can be written as:

[0085]

[0086] Among them, T refers to the number of leaf nodes, obj refers to the objective function, G j Refers to the first-order derivative of the objective function in the Taylor expansion of the jth leaf node, H j Refers to the second-order derivative of the Taylor expansion of the objective function at the jth leaf node;

[0087] Step S4.2: Optimize the learning rate γ and regularization parameter λ in step S4.1:

[0088] Step S4.2.1, set the respective search space Φ and search step size μ of the learning rate γ and regularization parameter λ, and set them as follows:

[0089] γ=(γ start ,γ end ,μ γ )

[0090] λ=(λ start ,λ end ,μ λ )

[0091] Among them, γ start and γ end are the search space Φ of the learning rate γ respectively. γ The upper and lower boundaries of μ γ is the search step size of the learning rate γ; λ start and λ end are the search space Θ of the regularization parameter λ respectively. λ Upper and lower boundaries, μ λ is the search step size of the regularization parameter λ;

[0092] Step S4.2.2: Generate a two-dimensional search parameter group matrix H according to the set search space and search step size. S , defined as follows:

[0093]

[0094] Where p is an integer, q is an integer,

[0095] Step S4.2.3: For H in step S4.2.2 S For each parameter group in the XGBoost classification model, the average classification accuracy of each parameter group is evaluated, and the parameter group with the highest average classification accuracy is selected. If there is one parameter group with the highest average classification accuracy, this parameter group is the selected parameter group. If there are multiple parameter groups with the highest average classification accuracy, then select λ from these multiple parameter groups. start +qμ λ The smallest parameter group;

[0096] Step S4.2.4, λ in the parameter group selected in step S4.2.3 start +qμ λ , γ start +pμ γThat is, the optimal learning rate γ and regularization parameter λ of the XGBoost classification model corresponding to the labeled low-dimensional data set in step S3;

[0097] Step S5: input the network traffic data set to be tested into the extreme gradient boosting XGBoost classification model in step S4 to obtain the business type classification result.

[0098] In order to illustrate the effectiveness of the method proposed in the present invention, the present invention provides an example. The sample data is collected by Figure 1 The constructed system is obtained through the network packet capture tool Wireshark, and the main protocol types and business types obtained are HTTP protocol, NTP protocol, DNS protocol, QQ, WeChat, Tencent video, E-mail, etc. In addition, 8 key feature indicators are considered, namely timestamp, source IP, destination IP, length of data packet, intermediate protocol, source port number, destination port number and whether it contains ACK / SYN fields.

[0099] Step 1: Use Wireshark to collect some labeled network traffic data sets, clean the data sets and extract features, and remove duplicate samples and invalid samples of incomplete data. The specific representation of the data set features is as follows:

[0100] F=[f1,f2,f3…f d ]

[0101] Step 2: For the convenience of analysis, the historical data is normalized. The specific representation of normalization is as follows:

[0102]

[0103] in Refers to the normalized i-th key feature index. max(f i ) refers to the maximum value of the i-th key feature indicator.

[0104] Step 3: Use the principal component analysis feature dimensionality reduction method to simplify complex multidimensional features into low-dimensional features, that is, convert the original d-dimensional data set into a k-dimensional data set, thereby retaining some of the most important features of the data set. The specific method of PCA feature dimensionality reduction is as follows:

[0105] (1) Organize the original data set into a matrix X with d rows and n columns.

[0106] (2) Decentralize the original data set, that is, where x i For a sample.

[0107] (3) Calculate the covariance matrix of the decentered data set:

[0108]

[0109] (4) Find the eigenvalues ​​and corresponding eigenvectors of the covariance matrix.

[0110] (5) Arrange the obtained eigenvectors into a matrix row by row from top to bottom according to the size of the corresponding eigenvalues, and take the first k rows to form the matrix P.

[0111] (6)Y=PX is the data after dimension reduction to k dimensions.

[0112] Finally, the principal component analysis algorithm was used to select the six most relevant characteristic indicators for measuring business types, such as Figure 2 As shown, using this algorithm can reduce the complexity of subsequent model training.

[0113] Step 4: Divide the labeled data set into a training set and a test set, input them into the Extreme Gradient Boosting Model (XGBoost) for training, tune the parameters of the classifier, and test the model after parameter tuning. The objective function of the Extreme Gradient Boosting Model (XGBoost) is as follows:

[0114]

[0115] in:

[0116]

[0117]

[0118] Then a greedy algorithm is used to find the optimal split. The basic idea is to split the leaf nodes one at a time from the root node and select the split based on specific conditions for each possible split. XGBoost also has specific criteria to choose the best split. Substitute the predicted value into the loss function and get the minimum value of the loss function.

[0119]

[0120] Step 5: Input the unlabeled data set actually obtained into the XGBoost classification model for testing to obtain the business type recognition result.

[0121] The above description is only a specific implementation of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a technician familiar with the technical field within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention.

Claims

1. A business type identification method based on PCA and XGBoost fusion, characterized in that: The following steps are involved: Step S1: Collect a network traffic data set with labels, where the service types of the network traffic data set include HTTP, NTP, DNS, QQ, WeChat, video, and email; Step S2, performing data cleaning and feature extraction on the network traffic data set in step S1, thereby obtaining a network traffic data set containing multi-dimensional features; Step S3, using the principal component analysis feature dimensionality reduction method to simplify the multi-dimensional features in step S2 into low-dimensional features, and obtain a labeled low-dimensional data set; Step S4, selecting key performance indicators with high correlation that affect the measurement of business types from the labeled low-dimensional data set in step S3, the key performance indicators constitute a labeled data set, the data set is divided into a training set and a test set, the training set is input into the extreme gradient boosting XGBoost classification model for training, and the learning rate γ and regularization parameter λ of the XGBoost classification model are tuned by an improved parameter tuning method to obtain the most suitable learning rate γ and regularization parameter λ for the network traffic data set, and the XGBoost classification model after parameter tuning is tested to obtain a trained extreme gradient boosting XGBoost classification model; The method for tuning the learning rate γ and the regularization parameter λ is as follows: Step S4.1, the extreme gradient boosting XGBoost classification model is: Among them, T refers to the number of leaf nodes, obj refers to the objective function, G j Refers to the first-order derivative of the objective function in the Taylor expansion of the jth leaf node, H j Refers to the second-order derivative of the Taylor expansion of the objective function at the jth leaf node; Step S4.2: Optimize the learning rate γ and regularization parameter λ in step S4.1: Step S4.2.1, set the respective search space Φ and search step size μ of the learning rate γ and regularization parameter λ, and set them as follows: γ=(γ start ,c end ,m γ ) λ=(λ start ,l end ,m λ ) Among them, γ start and γ end are the search space Φ of the learning rate γ respectively. γ The upper and lower boundaries of μ γ is the search step size of the learning rate γ; λ start and λ end are the search space Θ of the regularization parameter λ respectively. λ Upper and lower boundaries, μ λ is the search step size of the regularization parameter λ; Step S4.2.2: Generate a two-dimensional search parameter group matrix H according to the set search space and search step size. S , defined as follows: Where p is an integer, q is an integer, Step S4.2.3: For H in step S4.2.2 S For each parameter group in the XGBoost classification model, the average classification accuracy of each parameter group is evaluated, and the parameter group with the highest average classification accuracy is selected. If there is one parameter group with the highest average classification accuracy, this parameter group is the selected parameter group. If there are multiple parameter groups with the highest average classification accuracy, then select λ from these multiple parameter groups. start +qμ λ The smallest parameter group; Step S4.2.4, λ in the parameter group selected in step S4.2.3 start +qμ λ , γ start +pμ γ That is, the optimal learning rate γ and regularization parameter λ of the XGBoost classification model corresponding to the labeled low-dimensional data set in step S3; Step S5: input the network traffic data set to be tested into the extreme gradient boosting XGBoost classification model in step S4 to obtain the business type classification result.

2. According to the business type identification method based on PCA and XGBoost fusion according to claim 1, it is characterized in that: The multidimensional feature F in step S2 is specifically expressed as follows: <h2 style=";text-align:left;direction:ltr">F=[f1,f2,f3…f<h2 style=";text-align:left;direction:ltr"> d <h2 style=";text-align:left;direction:ltr"> ] F contains a vector of d features, f i represents the i-th key characteristic index, d≥i≥1, for f i The maximum values ​​of are normalized respectively, and the specific processing method is as follows: max(f i ) is the maximum value of the i-th key feature indicator, It is the i-th key feature index after normalization.

3. According to the business type identification method based on PCA and XGBoost fusion according to claim 1, it is characterized in that: The multi-dimensional features are simplified into low-dimensional features by using the principal component analysis feature dimensionality reduction method in step S3. Assuming that the original multi-dimensional features are d-dimensional, the simplified low-dimensional features are k-dimensional, where k < d; Step S3.1, the network traffic data set containing multi-dimensional features in step S2 is formed into a matrix X with d rows and n columns, wherein the network traffic data set containing multi-dimensional features has n samples, and each sample has d-dimensional features; Step S3.2, performing a decentralized process on the matrix X of d rows and n columns in step S2 to obtain a decentralized matrix X'; Step S3.3, calculate the covariance matrix Cov of the matrix X' in step S3.2: Among them, the superscript T' refers to the transpose; Step S3.4, find the eigenvalues ​​and corresponding eigenvectors of the covariance matrix; Step S3.5, arrange the obtained eigenvectors into a matrix row by row from top to bottom according to the size of the corresponding eigenvalues, and take the first k rows to form the matrix P; Step S3.6, Y=PX, where Y is a low-dimensional feature after being reduced to k dimensions by the principal component analysis feature dimensionality reduction method.

4. The business type identification method based on PCA and XGBoost fusion according to claim 1 is characterized in that: The actual data set without labels is obtained through the network packet capture tool Wireshark and input into the extreme gradient boosting model XGBoost classification model after PCA dimensionality reduction to obtain the business type classification result.

5. The business type identification method based on PCA and XGBoost fusion according to claim 1 is characterized in that: In step S4, the key performance indicators with high correlation include source IP, destination IP, source port number, destination port number, protocol type and data length.

Citation Information

Patent Citations

  • Network traffic protocol recognition method based on deep learning

    CN107682216A

  • Fault diagnosis and prediction method for small sample feature-oriented heterogeneous wireless network

    CN111586728A

  • Network traffic classification method and system, and electronic device

    WO2020062390A1

  • Network traffic classification method and system based on deep learning, and electronic device

    WO2020119481A1