A suspicious sample analysis and processing method and system
By obtaining the behavioral characteristics of malicious samples in the sandbox, selecting appropriate search starting points and strategies, and establishing feature inspection rules, the problem of low detection accuracy in the existing technology is solved, and more accurate sample analysis is achieved.
Patent Information
- Application Number
- CN202310980448.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-08-04
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2043-08-04
AI Technical Summary
In the prior art, the behavioral characteristics of samples in the sandbox are fixed and redundant, resulting in low detection accuracy and cannot meet the needs of complex sample analysis.
By obtaining all the behavioral characteristics of malicious samples in the sandbox, determining the search starting point and strategy, performing feature selection, and establishing feature testing rules to determine whether the sample to be detected is a malicious sample. The specific method includes determining the feature dimensions and subsets based on the number of initial feature sets, selecting an effective feature subset using different search strategies, and dynamically updating the threshold to improve the accuracy of feature selection.
The accuracy of sample detection and feature selection are improved, the detection accuracy is improved, and the feature identification is adaptively selected to detect unknown samples.
Smart Images

Figure CN117251691B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of sample analysis, and more specifically, to a suspicious sample analysis and processing method and system. Background Art
[0002] As a relatively safe dynamic analysis environment, sandbox technology can automatically analyze unknown samples and record the various behaviors of samples in the sandbox environment, so as to determine the true intention of the samples. At the same time, the operation of malicious samples in the sandbox will not cause any modifications to the real environment, which is a relatively ideal analysis environment.
[0003] In the prior art, whether a sample is suspicious is generally determined by analyzing its behavioral characteristics in a sandbox. However, the behavioral characteristics used for inspection are often relatively fixed and cannot meet the complex requirements of sample analysis. In addition, many behavioral characteristics are redundant, resulting in low detection accuracy and poor results.
[0004] Therefore, how to improve the accuracy of feature selection and detection precision is a technical problem that needs to be solved. Summary of the invention
[0005] The present invention provides a suspicious sample analysis and processing method to solve the technical problems of inaccurate behavior characteristics and low detection accuracy in the prior art. The method is applied in a sandbox and comprises:
[0006] Obtain multiple malicious samples and run them in sandboxes respectively;
[0007] Obtain all behavioral features of malicious samples in the sandbox and use them as the initial feature set;
[0008] Determine a search starting point and a search strategy based on the initial feature set, and perform feature selection according to the search starting point and the search strategy to obtain a valid feature subset;
[0009] Establish feature verification rules based on the valid feature subset, obtain the sample to be detected, and put the sample to be detected into the sandbox for operation. According to the feature verification rules, check the behavioral characteristics of the sample to be detected to determine whether the sample to be detected is a malicious sample.
[0010] In some embodiments of the present application, determining the search starting point and search strategy based on the initial feature set includes:
[0011] The feature dimension is determined according to the number of features in the initial feature set, the number of feature subsets is determined based on the feature dimension, and the search starting point and search strategy are determined based on the number of feature subsets.
[0012] In some embodiments of the present application, determining the search starting point and search strategy based on the number of feature subsets includes:
[0013] If the number of feature subsets does not exceed the first threshold, the search starting point is the entire set or the empty set, and feature selection is performed based on the search starting point through the first search strategy to obtain a valid feature subset;
[0014] If the number of feature subsets exceeds the first threshold value and does not exceed the second threshold value, a partial set is determined according to the excess amount of the first feature subset, the partial set is used as the search starting point, and feature selection is performed through the second search strategy based on the search starting point to obtain a valid feature subset;
[0015] If the number of feature subsets exceeds the second threshold, a partial set is determined according to the excess amount of the second feature subset, the partial set is used as a search starting point, and feature selection is performed based on the search starting point through a third search strategy to obtain a valid feature subset;
[0016] Among them, the first threshold is less than the second threshold, the first feature subset excess is the difference between the number of feature subsets and the first threshold, the second feature subset excess is the difference between the number of feature subsets and the second threshold, the full set is all behavioral features, and the empty set is no behavioral features.
[0017] In some embodiments of the present application, the first search strategy includes:
[0018] Taking the empty set or the full set as the search starting point, features are added or reduced according to the first evaluation rule;
[0019] When the first stop condition is met, the search is stopped and a valid feature subset is obtained;
[0020] The first stopping condition is that all feature subsets have been searched.
[0021] In some embodiments of the present application, the second search strategy includes:
[0022] Conduct the first search, which includes:
[0023] Obtaining attributes of the features according to the feature subset, determining performance energy of the features based on the attributes of the features, determining a first search number based on the performance energy of the features, and using the first search number as a second stop condition;
[0024] Determine the number of features to be added for each search based on the first search number and the feature subset, and perform searches in sequence;
[0025] Each search adds a corresponding number of features according to the second evaluation rule, determines a screening rate based on the feature dimension and the number of added features, screens the added features based on the screening rate, and adds the screened features to the partial set;
[0026] When the second stop condition is met, the search stops, and the first valid feature subset is obtained and used as the valid feature subset.
[0027] In some embodiments of the present application, the third search strategy includes:
[0028] Perform the first and second searches separately;
[0029] The second search includes:
[0030] Acquire information entropy and feature attributes of the feature, select attributes associated with the information entropy from the feature attributes, and obtain a proofreading amount, update the information entropy in real time based on the proofreading amount, set a second search number according to the information entropy of the feature, and use the second search number as a third termination condition;
[0031] Determine the random number based on the feature dimension, randomly select a random number in the partial set as the search starting point, and randomly increase the features in units of random numbers through the third evaluation rule;
[0032] When the third stop condition is met, the search stops and a second valid feature subset is obtained;
[0033] The first valid feature subset and the second valid feature subset are compared to obtain a feature difference set, and the feature difference set is added to the first valid feature subset as the valid feature subset.
[0034] In some embodiments of the present application, a sandbox key behavior is obtained, the sandbox key behavior is scored to obtain a sandbox score, and the first threshold and the second threshold are dynamically updated according to the sandbox score.
[0035] Correspondingly, the present application also provides a suspicious sample analysis and processing system, which is applied in a sandbox, and the system includes:
[0036] The test module is used to obtain a variety of malicious samples and put them into sandboxes for execution;
[0037] The acquisition module is used to obtain all behavioral features of malicious samples in the sandbox and use them as the initial feature set;
[0038] A selection module is used to determine a search starting point and a search strategy based on an initial feature set, and perform feature selection according to the search starting point and the search strategy to obtain a valid feature subset;
[0039] The detection module is used to establish feature verification rules based on the valid feature subset, obtain the sample to be detected, put the sample to be detected into the sandbox for operation, and verify the behavioral characteristics of the sample to be detected according to the feature verification rules to determine whether the sample to be detected is a malicious sample.
[0040] By applying the above technical solutions, a variety of malicious samples are obtained, and the malicious samples are put into the sandbox for operation respectively; all the behavioral features of the malicious samples in the sandbox are obtained and used as the initial feature set; the search starting point and search strategy are determined based on the initial feature set, and feature selection is performed according to the search starting point and search strategy to obtain a valid feature subset; feature verification rules are established according to the valid feature subset, and the samples to be detected are obtained, and the samples to be detected are put into the sandbox for operation, and the behavioral features of the samples to be detected are verified according to the feature verification rules to determine whether the samples to be detected are malicious samples. This application improves the accuracy of sample detection by reasonably selecting the behavioral features of malicious samples and establishing corresponding feature verification rules. Different search starting points and strategies are selected according to the dimensional adaptability of the behavioral features to obtain accurate feature identification for detecting unknown samples, thereby improving feature accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.
[0042] Figure 1 A schematic diagram showing a flow chart of a suspicious sample analysis and processing method proposed in an embodiment of the present invention;
[0043] Figure 2 A schematic diagram of the structure of a suspicious sample analysis and processing system proposed in an embodiment of the present invention is shown. DETAILED DESCRIPTION
[0044] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0045] The present application embodiment provides a suspicious sample analysis and processing method, which is applied in a sandbox, such as Figure 1 As shown, the method comprises the following steps:
[0046] Step S101, obtain multiple malicious samples, and put the malicious samples into sandboxes for execution.
[0047] In this embodiment, a variety of malicious samples are put into a sandbox for execution, and corresponding action execution logs are obtained to select features.
[0048] Step S102, obtaining all behavioral features of the malicious sample in the sandbox and using them as an initial feature set.
[0049] Step S103, determining a search starting point and a search strategy based on the initial feature set, and performing feature selection according to the search starting point and the search strategy to obtain a valid feature subset.
[0050] In this embodiment, different initial feature sets use different search starting points and strategies. The search starting point here refers to the basis for searching features, and the search strategy is to use a specific method to perform retrieval based on the search starting point to complete the selection of all features.
[0051] In order to improve the accuracy of feature selection, in some embodiments of the present application, a search starting point and a search strategy are determined based on an initial feature set, including: determining a feature dimension based on the number of features in the initial feature set, determining the number of feature subsets based on the feature dimension, and determining a search starting point and a search strategy based on the number of feature subsets.
[0052] In this embodiment, there is a certain relationship between the dimension and the feature subset. If the feature dimension is n, there are 2 n The number of feature subsets is different, and different search starting points and strategies are used to ensure the accuracy of feature selection.
[0053] In order to further improve the accuracy of feature selection, in some embodiments of the present application, the search starting point and search strategy are determined based on the number of feature subsets, including: if the number of feature subsets does not exceed the first threshold, the search starting point is the full set or the empty set, and feature selection is performed on the basis of the search starting point through the first search strategy to obtain a valid feature subset; if the number of feature subsets exceeds the first threshold but does not exceed the second threshold, a partial set is determined according to the excess amount of the first feature subset, the partial set is used as the search starting point, and feature selection is performed on the basis of the search starting point through the second search strategy to obtain a valid feature subset; if the number of feature subsets exceeds the second threshold, a partial set is determined according to the excess amount of the second feature subset, the partial set is used as the search starting point, and feature selection is performed on the basis of the search starting point through the third search strategy to obtain a valid feature subset; wherein the first threshold is less than the second threshold, the first feature subset excess amount is the difference between the number of feature subsets and the first threshold, the second feature subset excess amount is the difference between the number of feature subsets and the second threshold, the full set is all behavioral features, and the empty set is no behavioral features.
[0054] In this embodiment, when the number of feature subsets is small, the first search strategy, i.e., global search, is adopted to search all feature subsets, and the starting point is the full set or the empty set. When the number of feature subsets is large, the second search strategy (first search) is adopted, i.e., specific sequence search, and the starting point is the partial set. When the number of feature subsets is very large, the third search strategy is adopted, i.e., the combination of the first search and the second search (specific random search), and the starting point is also the partial set, which is different from the partial set in the second search strategy.
[0055] It should be noted that when there are fewer feature subsets, global search can ensure that the selected feature results are the most accurate and the time consumption is not too long, which is the global optimal solution. When there are more feature subsets, the first search (specific sequence search) is used, which has the lowest time complexity and the selected features are locally optimal. When there are many feature subsets, the first search and the second search are combined to select features, because the first search is only a local optimal solution, but it cannot guarantee a global optimal solution. Therefore, the two searches are combined to take into account both the local optimal solution and the global optimal solution.
[0056] In some embodiments of the present application, the first search strategy includes: taking the empty set or the full set as the search starting point, adding or reducing features according to the first evaluation rule; when the first stopping condition is met, stopping the search to obtain a valid feature subset; wherein the first stopping condition is that the entire feature subset search is completed.
[0057] In this embodiment, the first evaluation rule includes exhaustive method and branch and bound method, etc., to add or reduce features and complete the search of all feature subsets.
[0058] In some embodiments of the present application, the second search strategy includes: performing a first search, including: obtaining attributes of features according to a feature subset, determining the performance energy of the features based on the attributes of the features, determining a first search number based on the performance energy of the features, and using the first search number as a second stop condition; determining the number of features to be added for each search according to the first search number and the feature subset, and performing searches in sequence; adding a corresponding number of features according to a second evaluation rule for each search, determining a screening rate based on the feature dimension and the number of features added, screening the added features based on the screening rate, and adding the screened features to the partial set; when the second stop condition is met, the search stops, and a first valid feature subset is obtained and used as the valid feature subset.
[0059] In this embodiment, the second search strategy only performs the first search, and the attributes of the features include the correlation between features, the degree to which the features represent malicious samples, the degree of contradiction between features, and other attributes that represent the performance of feature subsets.
[0060] The formula for determining the performance energy of a feature based on its properties is as follows:
[0061]
[0062] Among them, P is the characteristic performance energy, is the weight of the i-th attribute, is the i-th attribute value, and n represents the number of attributes. n may be different for different malicious samples.
[0063] The performance energy of the feature is divided into multiple interval values, each interval value corresponds to a different first search number, and the first search number is used as the second stop condition. The number of features added for each search is determined based on the number of feature subsets and the first search number. The second evaluation rule includes local optimal algorithms such as greedy algorithms, and the optimal solution is selected each time. The features with the highest scores are added in sequence. In order to screen the features added for each search, the screening rate is determined based on the feature dimension and the number of features added. Each feature dimension and the number of features added have a corresponding screening rate. The features with the lowest scores among the added features are deleted according to the screening rate.
[0064] In some embodiments of the present application, the third search strategy includes: performing the first search and the second search respectively; the second search includes: obtaining the information entropy and feature attributes of the feature, screening out attributes associated with the information entropy from the feature attributes, and obtaining the proofreading amount, updating the information entropy in real time based on the proofreading amount, setting the second search number according to the information entropy of the feature, and using the second search number as the third termination condition; determining the random number based on the feature dimension, randomly selecting a random number in the partial set as the search starting point, and randomly adding features in units of random numbers according to the third judgment rule; when the third stop condition is met, the search stops and a second valid feature subset is obtained; comparing the first valid feature subset and the second valid feature subset to obtain a feature difference set, adding the feature difference set to the first valid feature subset, and using it as the valid feature subset.
[0065] In this embodiment, the third search strategy is to perform the first search and the second search respectively. The information entropy of the feature reflects the uncertainty and randomness of the system. Other attributes will also affect the information entropy and need to be updated.
[0066]
[0067] Among them, L is the feature correction amount, is the weight of the i-th attribute, is the i-th attribute value, and n represents the number of attributes related to information entropy.
[0068] The information entropy is updated and corrected according to the proofreading amount, and the information entropy of the feature is set to T. The updated information entropy is T*λL, where λ is the coefficient of the proofreading amount. Each information entropy interval value corresponds to a specific second search number.
[0069] The random number is determined based on the feature dimension. Each feature dimension interval value corresponds to a random number. A random number is randomly selected in the partial set as the search starting point. For example, if the partial set is 1 / 10 of the entire set and the random number is 2 features, then 2 features are randomly selected in the 1 / 10 entire set as the search starting point to jump out of the local optimum. Features are randomly added in units of random numbers through the third evaluation rule. The third evaluation rule includes random algorithms such as ant colony algorithm and genetic algorithm.
[0070] The first effective feature subset and the second effective feature subset are compared to obtain a feature difference set, and the feature difference set is added to the first effective feature subset. The feature difference set is a set of features that are present in the second effective feature subset but not in the first effective feature subset.
[0071] In order to improve detection accuracy, in some embodiments of the present application, sandbox key behaviors are obtained, the sandbox key behaviors are scored to obtain a sandbox score, and the first threshold and the second threshold are dynamically updated according to the sandbox score.
[0072] In this embodiment, the key sandbox behaviors include file operations, network communications, process operations, etc. Malicious samples may evade sandbox detection by detecting the above behaviors, and the first threshold and the second threshold need to be dynamically updated.
[0073]
[0074] K is the sandbox score, is the weight of the i-th type of behavior, is the behavior value of the i-th category, and n represents the number of behavior categories.
[0075] Assume that the first threshold and the second threshold are M1 and M2 respectively, then the updated thresholds are M1=α1*K*M01, α1 is the first update coefficient, M01 is the initial first threshold, and M2=α2*K*M02, α2 is the second update coefficient, and M02 is the initial second threshold.
[0076] Step S104, establish feature verification rules based on the valid feature subset, obtain the sample to be detected, put the sample to be detected into a sandbox for execution, and verify the behavior characteristics of the sample to be detected according to the feature verification rules to determine whether the sample to be detected is a malicious sample.
[0077] By applying the above technical solutions, a variety of malicious samples are obtained, and the malicious samples are put into the sandbox for operation respectively; all the behavioral features of the malicious samples in the sandbox are obtained and used as the initial feature set; the search starting point and search strategy are determined based on the initial feature set, and feature selection is performed according to the search starting point and search strategy to obtain a valid feature subset; feature verification rules are established according to the valid feature subset, and the samples to be detected are obtained, and the samples to be detected are put into the sandbox for operation, and the behavioral features of the samples to be detected are verified according to the feature verification rules to determine whether the samples to be detected are malicious samples. This application improves the accuracy of sample detection by reasonably selecting the behavioral features of malicious samples and establishing corresponding feature verification rules. Different search starting points and strategies are selected according to the dimensional adaptability of the behavioral features to obtain accurate feature identification for detecting unknown samples, thereby improving feature accuracy.
[0078] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present invention can be implemented by hardware, or by software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each implementation scenario of the present invention.
[0079] In order to further explain the technical idea of the present invention, the technical solution of the present invention is now described in combination with specific application scenarios.
[0080] Correspondingly, the present application also provides a suspicious sample analysis and processing system, which is applied in a sandbox, such as Figure 2 As shown, the system includes:
[0081] The testing module 201 is used to obtain multiple malicious samples and put the malicious samples into sandboxes for execution;
[0082] An acquisition module 202 is used to acquire all behavioral features of the malicious sample in the sandbox and use them as an initial feature set;
[0083] A selection module 203 is used to determine a search starting point and a search strategy based on the initial feature set, and perform feature selection according to the search starting point and the search strategy to obtain a valid feature subset;
[0084] The detection module 204 is used to establish feature verification rules based on the valid feature subset, obtain the sample to be detected, put the sample to be detected into the sandbox for execution, and verify the behavioral characteristics of the sample to be detected according to the feature verification rules to determine whether the sample to be detected is a malicious sample.
[0085] In some embodiments of the present application, the selection module 203 is used to:
[0086] The feature dimension is determined according to the number of features in the initial feature set, the number of feature subsets is determined based on the feature dimension, and the search starting point and search strategy are determined based on the number of feature subsets.
[0087] In some embodiments of the present application, the selection module 203 is used to:
[0088] If the number of feature subsets does not exceed the first threshold, the search starting point is the entire set or the empty set, and feature selection is performed based on the search starting point through the first search strategy to obtain a valid feature subset;
[0089] If the number of feature subsets exceeds the first threshold value and does not exceed the second threshold value, a partial set is determined according to the excess amount of the first feature subset, the partial set is used as the search starting point, and feature selection is performed through the second search strategy based on the search starting point to obtain a valid feature subset;
[0090] If the number of feature subsets exceeds the second threshold, a partial set is determined according to the excess amount of the second feature subset, the partial set is used as a search starting point, and feature selection is performed based on the search starting point through a third search strategy to obtain a valid feature subset;
[0091] Among them, the first threshold is less than the second threshold, the first feature subset excess is the difference between the number of feature subsets and the first threshold, the second feature subset excess is the difference between the number of feature subsets and the second threshold, the full set is all behavioral features, and the empty set is no behavioral features.
[0092] In some embodiments of the present application, the selection module 203 is used to:
[0093] Taking the empty set or the full set as the search starting point, features are added or reduced according to the first evaluation rule;
[0094] When the first stop condition is met, the search is stopped and a valid feature subset is obtained;
[0095] The first stopping condition is that all feature subsets have been searched.
[0096] In some embodiments of the present application, the selection module 203 is used to:
[0097] Conduct the first search, which includes:
[0098] Obtaining attributes of the features according to the feature subset, determining performance energy of the features based on the attributes of the features, determining a first search number based on the performance energy of the features, and using the first search number as a second stop condition;
[0099] Determine the number of features to be added for each search based on the first search number and the feature subset, and perform searches in sequence;
[0100] Each search adds a corresponding number of features according to the second evaluation rule, determines a screening rate based on the feature dimension and the number of added features, screens the added features based on the screening rate, and adds the screened features to the partial set;
[0101] When the second stop condition is met, the search stops, and the first valid feature subset is obtained and used as the valid feature subset.
[0102] In some embodiments of the present application, the selection module 203 is used to:
[0103] Perform the first and second searches separately;
[0104] The second search includes:
[0105] Acquire information entropy and feature attributes of the feature, select attributes associated with the information entropy from the feature attributes, and obtain a proofreading amount, update the information entropy in real time based on the proofreading amount, set a second search number according to the information entropy of the feature, and use the second search number as a third termination condition;
[0106] Determine the random number based on the feature dimension, randomly select a random number in the partial set as the search starting point, and randomly increase the features in units of random numbers through the third evaluation rule;
[0107] When the third stop condition is met, the search stops and a second valid feature subset is obtained;
[0108] The first valid feature subset and the second valid feature subset are compared to obtain a feature difference set, and the feature difference set is added to the first valid feature subset as the valid feature subset.
[0109] In some embodiments of the present application, a first module is further included, which is used to:
[0110] A sandbox key behavior is obtained, the sandbox key behavior is scored to obtain a sandbox score, and a first threshold and a second threshold are dynamically updated according to the sandbox score.
[0111] Those skilled in the art will appreciate that the modules in the system in the implementation scenario can be distributed in the system of the implementation scenario according to the implementation scenario description, or can be changed accordingly and located in one or more systems different from the implementation scenario. The modules in the above implementation scenario can be combined into one module, or can be further split into multiple submodules.
[0112] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.
Claims
1. A suspicious sample analysis and processing method, applied in a sandbox, characterized in that: The method comprises: Obtain multiple malicious samples and run them in sandboxes respectively; Obtain all behavioral features of malicious samples in the sandbox and use them as the initial feature set; Determine a search starting point and a search strategy based on the initial feature set, and perform feature selection according to the search starting point and the search strategy to obtain a valid feature subset; Establish feature verification rules based on the valid feature subset, obtain the sample to be tested, put the sample to be tested into the sandbox for operation, and verify the behavior characteristics of the sample to be tested based on the feature verification rules to determine whether the sample to be tested is a malicious sample; Determine the search starting point and search strategy based on the initial feature set, including: Determine the feature dimension based on the number of features in the initial feature set, determine the number of feature subsets based on the feature dimension, and determine the search starting point and search strategy based on the number of feature subsets; Determine the search starting point and search strategy based on the number of feature subsets, including: If the number of feature subsets does not exceed the first threshold, the search starting point is the entire set or the empty set, and feature selection is performed based on the search starting point through the first search strategy to obtain a valid feature subset; If the number of feature subsets exceeds the first threshold value and does not exceed the second threshold value, a partial set is determined according to the excess amount of the first feature subset, the partial set is used as the search starting point, and feature selection is performed through the second search strategy based on the search starting point to obtain a valid feature subset; If the number of feature subsets exceeds the second threshold, a partial set is determined according to the excess amount of the second feature subset, the partial set is used as a search starting point, and feature selection is performed based on the search starting point through a third search strategy to obtain a valid feature subset; Among them, the first threshold is less than the second threshold, the first feature subset excess is the difference between the number of feature subsets and the first threshold, the second feature subset excess is the difference between the number of feature subsets and the second threshold, the full set is all behavioral features, and the empty set is no behavioral features; The first search strategy includes: Taking the empty set or the full set as the search starting point, features are added or reduced according to the first evaluation rule; When the first stop condition is met, the search is stopped and a valid feature subset is obtained; Among them, the first stop condition is that all feature subsets are searched; The second search strategy included: Conduct the first search, which includes: Obtaining attributes of the features according to the feature subset, determining performance energy of the features based on the attributes of the features, determining a first search number based on the performance energy of the features, and using the first search number as a second stop condition; Determine the number of features to be added for each search based on the first search number and the feature subset, and perform searches in sequence; Each search adds a corresponding number of features according to the second evaluation rule, determines a screening rate based on the feature dimension and the number of added features, screens the added features based on the screening rate, and adds the screened features to the partial set; When the second stop condition is met, the search stops, and the first valid feature subset is obtained and used as the valid feature subset; The third search strategy includes: Perform the first and second searches separately; The second search includes: Acquire information entropy and feature attributes of the feature, select attributes associated with the information entropy from the feature attributes, and obtain a proofreading amount, update the information entropy in real time based on the proofreading amount, set a second search number according to the information entropy of the feature, and use the second search number as a third termination condition; Determine the random number based on the feature dimension, randomly select a random number in the partial set as the search starting point, and randomly increase the features in units of random numbers through the third evaluation rule; When the third stop condition is met, the search stops and a second valid feature subset is obtained; Compare the first valid feature subset with the second valid feature subset to obtain a feature difference set, add the feature difference set to the first valid feature subset, and use it as the valid feature subset; A sandbox key behavior is obtained, the sandbox key behavior is scored to obtain a sandbox score, and a first threshold and a second threshold are dynamically updated according to the sandbox score.
2. A suspicious sample analysis and processing system, characterized in that: The method for analyzing and processing suspicious samples according to claim 1, wherein the system comprises: The test module is used to obtain a variety of malicious samples and put them into sandboxes for execution; The acquisition module is used to obtain all behavioral features of malicious samples in the sandbox and use them as the initial feature set; A selection module is used to determine a search starting point and a search strategy based on an initial feature set, and perform feature selection according to the search starting point and the search strategy to obtain a valid feature subset; The detection module is used to establish feature verification rules based on the valid feature subset, obtain the sample to be detected, put the sample to be detected into the sandbox for operation, and verify the behavioral characteristics of the sample to be detected according to the feature verification rules to determine whether the sample to be detected is a malicious sample.
