Efficient DDoS attack detection method based on interpretable AI
Through the CapNet model, combined with one-dimensional ResNet50, SelfNet and Captum modules, the problems of gradient vanishing, robustness and interpretability in DDoS attack detection are solved, and efficient and accurate DDoS attack detection is achieved, suitable for complex network traffic.
Patent Information
- Application Number
- CN202411377622.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-30
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2044-09-30
AI Technical Summary
The existing technology is difficult to quickly and accurately identify and detect DDoS attacks. Traditional methods rely on manual analysis and detection systems based on attack signatures are invalid for new DDoS attacks. The deep learning model has gradient vanishing problems, insufficient robustness, lack of interpretation and high computational costs. .
A highly efficient DDoS attack detection method based on interpretable AI is proposed. Combined with one-dimensional ResNet50, SelfNet and Captum modules, the gradient vanishing problem is optimized, robustness and interpretability are improved, and detection is improved through feature extraction and feature contribution analysis. Efficiency and precision.
CapNet significantly improves the speed and accuracy of DDoS attack detection, reduces computing resource consumption, enhances the robustness and interpretability of the model, and maintains efficient and high accuracy in large-scale data processing, and is suitable for the detection of complex network traffic.
Smart Images

Figure FT_1 
Figure FT_2 
Figure SMS_1
Abstract
Description
Technical Field
[0001] The present invention relates to the field of network security, and in particular to an efficient DDoS attack detection method based on explainable AI. Background Art
[0002] Digital transformation has significantly changed people's lifestyles, but it has also brought new security challenges. Distributed denial of service (DDoS) attacks have become one of the most destructive threats in cyberspace. Such attacks send a large amount of traffic to the target server or network, making it unable to provide services to legitimate users. According to statistics, about 7.9 million DDoS attacks were recorded worldwide in the first half of 2023, which not only paralyzed services, but also seriously affected users' quality of service (QoS).
[0003] DDoS attacks are characterized by significant distribution, suddenness, and concealment, which poses a huge challenge to network traffic analysis. Attackers usually forge and proxy source addresses and use "zombie computers" around the world to launch attacks, making the source addresses widely distributed and difficult to track. The attack traffic surges in a very short period of time, causing a huge impact on the network; at the same time, the attack traffic is disguised as normal traffic and is extremely concealed. These characteristics make DDoS attacks difficult to prevent and detect.
[0004] Currently, many Internet companies and cloud service providers invest heavily every year to deal with DDoS attacks. However, the method of relying on manual analysis of massive network data can no longer meet the needs of real-time detection. Traditional attack signature-based detection systems rely on manual extraction of features of known attacks and are ineffective against new DDoS attacks. Therefore, how to quickly and accurately identify DDoS attacks from complex network traffic and achieve refined analysis and graphical display of traffic has become an urgent problem to be solved in the field of network security.
[0005] Over the years, machine learning technology has been widely used in DDoS attack detection. Traditional models such as K-nearest neighbor (KNN), support vector machine (SVM), random forest (RF) and naive Bayes (NB) identify anomalies by classifying network traffic patterns. However, as DDoS attack methods continue to evolve and become more complex, the detection efficiency of these methods has gradually declined.
[0006] In recent years, deep learning has provided a new research direction for DDoS attack detection. Its powerful data representation ability enables it to identify complex traffic patterns. Convolutional neural networks (CNNs) show good DDoS detection potential by parsing raw data patterns; deep autoencoders can create concise data representations and capture subtle malicious traffic patterns; artificial neural networks (ANNs) reveal the complex relationships between input data through nonlinear activation functions, providing deep insights for detecting potential threats.
[0007] Although deep learning has significantly improved the accuracy, real-time, and automation of DDoS attack detection, it still faces four key challenges. First, as the number of network layers increases, the gradient vanishing problem is prone to occur. Second, the change in data distribution in an open environment affects the robustness of the model. Third, complex deep learning models lack interpretability, making it difficult to track the causes of classification decisions. Finally, it is challenging to maintain high accuracy and efficiency when dealing with large models and complex features, which can easily lead to overfitting and increased computational costs. Summary of the invention
[0008] This paper proposes an efficient DDoS attack detection method based on explainable AI, CapNet, which is used to optimize the gradient vanishing problem in DDoS attack detection, improve robustness, enhance interpretability, and maintain high accuracy and efficiency. To achieve these goals, this paper adopts the following technical solutions. The system consists of: 1. One-dimensional ResNet50
[0009] The one-dimensional ResNet50 model in CapNet adapts the original ResNet50's two-dimensional convolutional layer and pooling layer to a one-dimensional structure to adapt it to the feature processing of network traffic, and retains the residual connection to avoid the gradient vanishing problem. At the same time, the modified fully connected layer can better handle the output of one-dimensional features, thereby improving the classification performance and efficiency of the model. These adjustments not only reduce the consumption of computing resources, but also ensure high accuracy and robustness when processing large-scale DDoS attack data. Secondly, the robustness of the model was verified through five-fold cross-validation and the introduction of new data sets.
[0010] 2.SelfNet module
[0011] Although ResNet50 already has powerful feature extraction capabilities, when processing complex network traffic data, relying solely on ResNet50 may not be enough to fully meet the requirements of specific tasks, especially when high-precision detection of DDoS attacks is required. Therefore, the SelfNet module is introduced to optimize feature extraction. Its main components include multiple fully connected layers, which further abstract and optimize the features extracted from ResNet50 through nonlinear mapping. The working principle of SelfNet is to stack fully connected layers layer by layer, perform feature conversion and activation function processing on each layer, and then capture complex feature relationships and high-level abstract information. This design enables the model to more effectively improve classification performance when processing complex network traffic, ensure the refinement of feature extraction, and thus enhance the overall performance of the model.
[0012] 3. Captum module
[0013] The main purpose of introducing the Captum module is to solve the problem of the "black box" characteristics of deep learning models, that is, the lack of explainability of the model's decision-making process. DDoS attack detection systems require clear explanations of the contribution of each feature so that network administrators can understand and trust the output of the model.
[0014] The Captum module is a key component in CapNet for enhancing model interpretability. Its main function is to make the model's decision-making process more transparent by quantifying the contribution of each input feature to the model output, making it easier for users to understand why the model makes a certain prediction. The core mechanism of Captum is to calculate the impact of each input feature on the final prediction result through methods such as the integrated gradient method.
[0015] The working principle of the Captum module can be summarized in the following steps:
[0016] S31. Select baseline input: First, select a baseline input as a reference point, usually a zero vector or other standard input. This baseline input serves as the starting point for feature contribution.
[0017] S32, interpolated input data: Captum generates multiple interpolation points from the baseline input to the actual input. These interpolation points represent the transition state from "no features" to "actual features".
[0018] S33. Calculate gradients: For each interpolation point, Captum calculates the gradient of the model output relative to the input features, quantifying the contribution of each feature to the model prediction.
[0019] S4, Cumulative Contribution: By integrating the gradients of all interpolation points, Captum finally generates the total contribution of each input feature to the model output.
[0020] The formula is as follows:
[0021]
[0022] in:
[0023] Official IG i (x) represents the input feature x i Contribution to the model prediction result f(x). i is the ith feature of the actual input, x′ i is the i-th feature of the baseline input, usually the zero vector or the minimum value of the input. The parameter α is an interpolation parameter ranging from 0 to 1. The expression Represents the partial derivative of the model output with respect to the input feature x_i at the point x′+α(xx′) on the interpolation path.
[0024] The steps include:
[0025] S1. Obtain the input DDoS traffic data set and pre-process the traffic data into a one-dimensional tensor;
[0026] S2. Input the one-dimensional tensor into the CapNet model to obtain the traffic classification results and feature contribution.
[0027] S3. Further select feature combinations based on feature contribution to achieve the purpose of efficiently detecting DDoS attacks.
[0028] Further: S1 comprises the following sub-steps:
[0029] S11, data loading: read the DDoS traffic dataset through the CSV file;
[0030] S12. Data type conversion and processing: Try to convert the feature column into numerical data, replace the part that cannot be converted with NaN to ensure data consistency, and delete the records containing missing values;
[0031] S13, feature normalization: normalize the data to ensure that the features have the same scale and reduce the impact on the model;
[0032] S14. Dataset division: Use the cross-validation method to divide the dataset into training set and test set to ensure the generalization performance of the model;
[0033] S15, Tensor conversion: Convert the preprocessed training data and test data into one-dimensional tensors to prepare for subsequent deep learning model input.
[0034] Further: S2 comprises the following sub-steps:
[0035] The traffic input data is subjected to feature extraction through several ResNet residual blocks, and finally the traffic classification result is obtained. Among them, the residual block uses a method combining local feature extraction with global feature extraction to extract multi-level features of the input traffic.
[0036] Furthermore: the local feature extraction module includes a first one-dimensional convolutional layer, a batch normalization layer (BatchNorm1d) and a ReLU activation function connected in sequence.
[0037] Furthermore: the global feature extraction module is a global average pooling layer (GlobalAverage Pooling), which is used to extract the global representation of local features and input it into the fully connected layer for classification.
[0038] The specific steps are as follows:
[0039] SA1, convolution and initial feature extraction: The one-dimensional tensor is input into the network, and the initial feature extraction is performed through the first one-dimensional convolutional layer (kernel size is 7, stride is 2), and after batch normalization (BatchNorm1d) and ReLU activation function, the initial feature map is output.
[0040] SA2, residual block processing: The initial feature map passes through multiple residual blocks, each of which consists of two layers of one-dimensional convolution, batch normalization, and ReLU activation function. The residual connection ensures that the input and output features can be superimposed, thereby effectively alleviating the gradient vanishing problem and enhancing the feature learning ability.
[0041] SA3, global feature extraction: After being processed by the residual block, the feature map passes through the global average pooling layer to compress all feature dimensions into a global feature representation of a fixed length.
[0042] SA4, classification result output: The global feature representation passes through the fully connected layer (FC layer) and is processed by several linear layers and ReLU activation layers to finally output the traffic classification result.
[0043] Further: In S3, feature combinations are further selected according to feature contribution to achieve the purpose of efficiently detecting DDoS attacks, and the specific method is as follows:
[0044] S31. Calculate feature contribution: Use the integrated gradient method to analyze the importance of input features through the Captum module and calculate the contribution of each feature to the model output. This method quantifies the impact of each feature on the model classification result by integrating the gradient of the model output relative to each input feature.
[0045] S32. Screen important features: According to the calculated feature contribution, select the feature combination with higher contribution, and retain the features that have a significant impact on DDoS attack detection. Low contribution or redundant features will be eliminated to improve the detection efficiency and accuracy of the model.
[0046] S33. Optimize feature set: By repeatedly iteratively adjusting feature combinations, verify the impact of different feature combinations on DDoS attack detection, and thus determine an optimal feature subset to ensure that the computational complexity of the model is reduced while retaining the detection accuracy.
[0047] S34, Final Detection: Use the selected feature combination to perform the final detection of the model, effectively improving the speed and accuracy of DDoS attack detection.
[0048] The beneficial effects of the present invention are:
[0049] (1) The present invention provides a fast and efficient DDoS attack detection method based on explainable AI - CapNet. By combining the optimized ResNet50 model with the SelfNet module, the detection speed and accuracy of DDoS attacks in network traffic are effectively improved. The CapNet model is optimized for one-dimensional traffic data, which greatly reduces the consumption of computing resources while improving the response speed of the model. The SelfNet module further enhances the effect of feature extraction, thereby improving the accuracy of classification. In order to increase the interpretability of the model, the Captum module is added to help users understand the decision-making process of the model by analyzing the contribution of each feature to the detection results. Compared with existing methods, the present invention significantly improves the detection efficiency and accuracy of DDoS attacks.
[0050] (2) The purpose of the present invention is to quickly and accurately detect DDoS attacks in the network. By combining the optimized ResNet50 structure and the SelfNet module, a detection method that can efficiently process one-dimensional network traffic data is proposed. The present invention not only improves the detection speed of DDoS attacks, but also improves the detection accuracy. Through the analysis of feature contribution by the Captum module, the interpretability of the model is further enhanced, helping users understand how the model derives the detection results, thereby achieving fast and reliable DDoS attack detection. BRIEF DESCRIPTION OF THE DRAWINGS
[0051] Figure 1 It is a model diagram of the present invention;
[0052] Figure 2 This is the SelfNet module structure diagram. DETAILED DESCRIPTION
[0053] This paper proposes a DDoS attack detection model CapNet based on explainable AI, which can efficiently and accurately detect DDoS attacks in the network by using deep learning technology and feature selection methods. It is characterized in that the model combines the preliminary feature screening of random forests and the feature contribution analysis of deep learning models, which can gradually reduce the number of features and maintain high detection accuracy while significantly reducing the computational complexity. The specific implementation steps are as follows:
[0054] S1, input data preprocessing;
[0055] S2. Input the data into the CapNet model to obtain the traffic classification results and feature contribution;
[0056] S3, further select feature combinations according to feature contribution to find the minimum feature combination to achieve the purpose of efficiently detecting DDoS attacks;
[0057] S4. Use the key feature set after feature screening to verify on other data sets;
[0058] The S1 comprises the following sub-steps:
[0059] S11. The input network traffic data contains 79-dimensional features. Missing values and invalid values are processed through data cleaning;
[0060] S12. Use feature scaling technology to standardize the data to ensure fairness between features;
[0061] S13. Convert the flow data into a one-dimensional tensor.
[0062] In S2, the CapNet model is based on the ResNet50 architecture, and the residual network is optimized to enable it to process one-dimensional traffic data. The model is trained using a preprocessed feature set, and the SelfNet module is used to further optimize the features extracted from ResNet50 to improve the accuracy of the model. The Captum module is used to calculate feature contributions, increase interpretability, and improve model transparency.
[0063] like Figure 1 As shown, in this embodiment, the CapNet model consists of optimized ResNet50, SelfNet and Captum modules, which are used to efficiently detect DDoS attacks in the network. First, the input 79-dimensional network traffic feature sequence is preliminarily processed by the optimized structure of ResNet50. The convolution layer of ResNet50 has been adjusted to a one-dimensional convolution layer (Conv1d), which can process one-dimensional traffic data characteristics and extract the initial time series features through a one-dimensional convolution kernel. Subsequently, the data stream passes through multiple bottleneck blocks (BottleneckBlock) in sequence, each of which contains three convolution operations and batch normalization (BatchNormalization). The residual connection ensures effective information transmission, avoids the disappearance of gradients, and fully extracts the key features in the time series. Through the adaptive average pooling layer (AdaptiveAvgPool1d), input sequences of different lengths are compressed into feature representations of uniform size to ensure the uniformity and stability of subsequent processing.
[0064] Then, the data after ResNet50 feature extraction enters the SelfNet module for deep optimization. Figure 2As shown in the figure, the SelfNet module consists of multiple layers of fully connected layers, which aims to further abstract and optimize the features extracted from ResNet50. First, the features are linearly transformed through the fully connected layer, and then batch normalization is used to maintain numerical stability to ensure that the data is not numerically distorted during the transmission between layers. Nonlinearity is introduced through the ReLU activation function to enhance the model's ability to express complex features. This process goes through multiple iterations, abstracting high-level feature information layer by layer, and finally generating the most optimized feature representation in the last fully connected layer. Through the processing of the SelfNet module, the model can more accurately identify complex patterns in network traffic, effectively improving the detection accuracy of DDoS attacks and the robustness of the model.
[0065] Finally, the Captum module is introduced for feature importance analysis. This module uses the Integrated Gradients method to calculate the contribution of each input feature to the model prediction result. By interpolating the input features with the baseline input, Captum calculates the gradient of the model output to the feature, and integrates the gradient of the interpolation point to finally obtain the contribution of each feature. Through Captum's feature analysis, the interpretability of the model has been significantly improved, allowing users to intuitively understand which features have a key impact on the final classification results, enhancing the transparency and credibility of the model. The model ultimately outputs the classification results of network traffic and the contribution of each feature, providing a reliable basis for efficient detection of DDoS attacks.
[0066] In S3, the Captum module uses the integrated gradient method to calculate the contribution of each feature to the model output. The specific calculation steps are as follows: First, select the baseline input as a reference (usually a full zero vector or a minimum value), and then interpolate between the actual input and the baseline input to generate multiple intermediate input points. For each interpolation point, calculate the gradient of the model output relative to each input feature. Finally, integrate these gradient values along the interpolation path to obtain the overall contribution of each feature to the model output. In this way, Captum can effectively quantify the impact of each input feature on the classification result, thereby helping to guide feature selection and find the minimum feature set to maintain the performance of efficient detection of DDoS attacks. S31, the F1 score of the 79 features initially selected after training is 99.96%. Through feature contribution analysis, 15 key features are selected, namely: Total Fwd Packets, Total Length of Bwd Packets, Fwd Packet Length Min, Bwd Packet Length Std, Bwd URG Flags, Bwd Packets / s, Max Packet Length, PSHFlag Count, ACK Flag Count, Average Packet Size, Avg Bwd Segment Size, BwdAvg BulkRate, Subflow Bwd Bytes, Init Winbytesbackward, Similar HTTP. Use these features to continue training the model.
[0067] S32. After further feature selection, the accuracy of the model is 99.81%. Four features are selected, namely: total number of forward packets (Total Fwd Packets), minimum length of forward packets (Fwd Packet Length Min), maximum packet length (Max Packet Length), and confirmation flag count (ACKFlag Count), and the model is trained based on these features.
[0068] S33. After the second round of feature screening, the F1 score of the model was 99.75%. Finally, the two most contributing features were selected: Maximum Packet Length and Acknowledgement Flag Count.
[0069] S34. Finally, the model was trained using these two features, and the F1 score reached 99.17%. Feature contribution analysis shows that these two features are representative in DDoS attack detection:
[0070] Maximum packet length: used to detect DDoS attacks. DDoS attacks often manifest as a flood of large numbers of small packets or abnormally large packets, resulting in excessive consumption of network resources. Therefore, the maximum packet length is an important feature for identifying attacks.
[0071] ACK Flag Count: used to distinguish normal traffic from abnormal traffic. The stability of TCP connection and the reliability of data transmission in normal traffic are reflected by the ACK flag. By monitoring the ACK flag count, normal traffic and abnormal traffic can be effectively distinguished.
[0072] By further reducing the number of features, the training time of the model is significantly reduced, from 246.77 seconds for 2 features to 2840.63 seconds for 79 features, a reduction of nearly 10 times.
[0073] In S4, the generalization ability and robustness of the model are further evaluated by applying the most representative features screened in the early stage to different data sets. During the verification process, the model can still maintain high-precision DDoS attack detection effects in other data sets, with an F1 score of up to 98.78%, proving the good adaptability and detection stability of this feature combination in different network environments. This verification step ensures the wide applicability of the model in diverse scenarios and can effectively respond to DDoS attack detection needs under different network structures and traffic patterns.
[0074] Due to the wide variety of DDoS attacks and the significant differences in traffic characteristics of different attack types, it is necessary to ensure that the model can not only detect a single type of DDoS attack, but also maintain efficient detection performance when facing multiple attack types. In order to verify the generalization ability and robustness of the CapNet model in dealing with multiple types of attacks, the present invention conducted a multi-classification experiment, covering a variety of common DDoS attack types, namely PORTMAP, LDAP, MSSQL, NetBIOS, SYN, UDP, and UDPLag. The specific steps are as follows: S1, data set preprocessing;
[0075] S2. Model training and classification.
[0076] The specific steps in S1 include:
[0077] S11. Obtain a network traffic data set containing various DDoS attack types, ensuring that the data set covers common attack types (PORTMAP, LDAP, MSSQL, NetBIOS, SYN, UDP, UDPLag) and normal traffic, and ensuring the diversity and representativeness of the data;
[0078] S12. Standardize the acquired traffic data so that each feature dimension has the same numerical range, ensure the dimensional consistency between features, and eliminate the impact caused by scale differences between features;
[0079] S13. Convert the standardized traffic data into a one-dimensional tensor suitable for deep learning model processing as input for subsequent model training and inference.
[0080] The specific steps in S2 include:
[0081] S21. Train the CapNet model on the preprocessed traffic data. First, perform preliminary feature extraction through the optimized ResNet50 network, use a one-dimensional convolutional layer to process the temporal characteristics of network traffic, and use residual connections to avoid gradient vanishing and extract deep-level features. Then, input the extracted features into the SelfNet module for further nonlinear feature mapping and high-dimensional feature optimization to enhance the classification performance of the model.
[0082] S22. Multiple experiments were conducted using different feature combinations (79 features, 15 features, 4 features, and 2 features) to evaluate the effects of each feature combination in multi-classification tasks, compare the effects of different combinations on classification accuracy, training time, and model robustness, and finally select the optimal feature subset to achieve efficient and accurate DDoS attack detection.
[0083] In the multi-classification experiment of the present invention, the CapNet model was tested on a variety of DDoS attack types (including PORTMAP, LDAP, MSSQL, NetBIOS, SYN, UDP, UDPLag). The experimental results show that different feature combinations have a significant impact on the classification performance of the model. When 79 features are used, the classification accuracy of the model in each attack type is close to 100%, such as PORTMAP (99.76%), LDAP (99.35%), MSSQL (99.75%), SYN (99.86%) and UDPLag (100%), indicating that the model has extremely high classification performance under all features. When 15 features are used, although the overall accuracy has decreased, the model can still maintain a high classification performance, such as the accuracy of PORTMAP (99.08%) and NetBIOS (98.95%) is close to 98% or more, but the accuracy of some types such as UDP (92.71%) has decreased. When using 4 features, the overall accuracy of the model decreased, especially for LDAP (94.82%) and UDP (88.41%) types, indicating that the reduction in the number of features has an impact on model performance. When using 2 features, the accuracy of the model is still high for some types, such as MSSQL (99.95%) and SYN (99.90%), but the accuracy of UDP (87.31%) and UDPLag (84.55%) types is significantly reduced, indicating that the over-simplification of features has a poor detection effect on some attack types.
[0084] In general, as the number of features decreases, the classification accuracy of the model will decrease, but with the combination of two features, the model can still maintain a high accuracy in most attack types.
[0085] The advantages of the present invention are as follows:
[0086] 1. Efficient DDoS attack detection: By combining the optimized ResNet50, SelfNet and Captum modules, it can accurately identify DDoS attack patterns in network traffic and effectively screen out the most representative features through the integrated gradient method, thereby maintaining high-precision detection effects while reducing the feature set.
[0087] 2. Significantly reduce training time: Through feature contribution analysis, the model training time is significantly reduced by gradually screening from 79 features to the final two key features. Compared with the complete feature set, the training time is reduced by nearly 90% after the feature set is reduced, which greatly improves the model training efficiency and computing resource utilization.
[0088] 3. Enhance the interpretability of the model: The Captum module calculates the contribution of each feature to the model output through the integrated gradient method, making the model highly interpretable. Users can clearly understand which features have the greatest impact on the classification results, improving the transparency and credibility of the model.
[0089] 4. Wide adaptability: The present invention demonstrates strong generalization ability and robustness on multiple data sets, and the detection accuracy is always maintained above 99%, which proves the wide adaptability of the model in different network scenarios and is suitable for DDoS attack detection in various complex network environments.
[0090] 5. Improvement of resource utilization: The present invention reduces the complexity of the model through feature screening and significantly reduces the consumption of computing resources. In the end, only two features are used to achieve the purpose of efficient detection, which greatly improves the real-time performance and resource utilization efficiency of the system, making the model have higher application potential in large-scale traffic monitoring. In summary: The present invention proposes an efficient DDoS attack detection model based on explainable AI, CapNet. By combining the optimized ResNet50, SelfNet module and Captum module, the model can significantly improve the detection efficiency of DDoS attacks in network traffic while ensuring detection accuracy, and enhance the interpretability of the model. By introducing one-dimensional convolution to extract features from the time series of traffic data, CapNet can effectively capture the characteristics of DDoS attacks. The SelfNet module further optimizes the features and improves the classification ability of the model in complex traffic. The Captum module calculates the feature contribution to enhance the transparency of the model and help users understand the decision-making process of the model. In multi-classification experiments, CapNet demonstrated superior detection capabilities for various types of DDoS attacks. Even when the number of features is reduced, the model can still maintain high detection accuracy and robustness, proving its efficiency and reliability in large-scale DDoS attack detection. The detection method provided by the present invention is not only applicable to binary classification tasks, but can also effectively cope with multi-type attack scenarios, has broad application prospects, and can provide an efficient, accurate, and explainable DDoS attack detection solution for network security.
Claims
1. A DDoS attack detection model CapNet based on explainable AI, It is characterized in that The following steps are involved: S1: Preprocess the network traffic data. S2: Convert the preprocessed traffic data into tensors and input them into the CapNet model for classification. S3: Calculate and output feature contribution.
2. According to the explainable AI-based DDoS attack detection model of claim 1, It is characterized in that The step S1 comprises the following sub-steps: S11: Remove missing values, replace infinite values with NaN, and fill missing values with column means to ensure data consistency. S12: Perform feature scaling and standardization, using Z-score standardization to convert the features into a distribution with a mean of 0 and a standard deviation of 1. S13: In high-speed network environments, system sampling technology is used to improve processing efficiency.
3. According to the explainable AI-based DDoS attack detection model of claim 1, It is characterized in that The step S2 comprises the following sub-steps: S21: Optimize the ResNet50 model to adapt it to one-dimensional data processing, reduce the number of network layers and convolution kernel size, reduce computational complexity and resource consumption, and use it for feature extraction and classification of traffic data. S22: Features are further optimized through the SelfNet module. SelfNet consists of multiple fully connected layers and is used for nonlinear mapping and high-level feature extraction to improve the classification performance of the model. S23: Perform feature importance analysis through the Captum module. Captum uses the integrated gradient method to calculate the contribution of each input feature to the model output, optimize the interpretability of the model, and shorten the detection time.
4. The DDoS attack detection model based on explainable AI according to claim 1, It is characterized in that The step S3 comprises the following sub-steps: Captum calculates feature contributions through the integrated gradient method. Captum first selects the baseline input, then generates interpolation points from the baseline to the actual input, gradually calculates the gradient of each interpolation point, and integrates all gradients to finally obtain the contribution of each feature, thereby quantifying its impact on the model output and improving the transparency and interpretability of the model.
5. The DDoS attack detection model based on explainable AI according to claim 4, It is characterized in that Based on the feature importance output by Captum, the features that contribute most to DDoS attack detection are further selected to optimize the detection effect.
Citation Information
Patent Citations
Internet of Things DDoS attack detection method and device based on space-time hybrid model
CN116886345A
Network intrusion analysis method based on interpretable artificial intelligence technology
CN117997624A
System and method for network intrusion detection
WO2023052838A1
Cited By
APP traffic security reinforcement method and system based on interpretable AI
CN121530700A
