2019年最新-BoundaryDetectioninTokenizingNetworkApplicationPayload:在網絡應用的有效載荷邊界檢_第1頁
2019年最新-BoundaryDetectioninTokenizingNetworkApplicationPayload:在網絡應用的有效載荷邊界檢_第2頁
2019年最新-BoundaryDetectioninTokenizingNetworkApplicationPayload:在網絡應用的有效載荷邊界檢_第3頁
已閱讀5頁,還剩19頁未讀, 繼續免費閱讀

下載本文檔

版權說明:本文檔由用戶提供并上傳,收益歸屬內容提供方,若內容存在侵權,請進行舉報或認領

文檔簡介

1、Boundary Detection in Tokenizing Network Application Payload for Anomaly DetectionRachna Vargiya and Philip ChanDepartment of Computer SciencesFlorida Institute of TechnologyMotivationExisting anomaly detection techniques rely on informatio n derived only from the packet headersMore sophisticated at

2、tacks involve the application payloadExample : Code Red II wormGET /default ida?NNNNNNNNNParsing the payload is required!Problems in hand-coded parsing:Large number of application protocolsFrequent introduction of new protocolsProblem StatementTo parse application payload into tokens without explici

3、t knowledge of the application protocolsThese tokens are later used as features for anomaly detectionRelated workPattern Detection Important TokensFixed Length:Forrest et al. (2019)Variable Length:Wespi et al. (2000)Jiang et al.(2019)Boundary Detection - All Tokenso VOTING EXPERTS by Cohen et al. (2

4、019)Boundary EntropyFrequencyBinary VotesApproachBoundary Finding Algorithms:o Boundary Entropyo FrequencyAugmented Expected Mutual InformationMinimum Description LengthApproach is domain independent (no prior domain knowledge)Combining Boundary FindingAlgorithmsCombination of all or a subset (E.g.

5、Frequency + Minimum Description Length) of techniquesEach algorithm can cast multiple votes, depending on confidence measureBoundary Entropy (Cohen et al)Entropy at the end of each possible window is calculatedHigh Entropy means more variationw XItis 1 rai ny dayis the byte following the current win

6、dowVoting using Boundary Entropy change graph to discrete bars )Itiseirainydayn n n n n n n n n o n n n nEntropy in meaningful tokens starts with a high value, drops, and peaks at the end Vote for positions with the peak entropy Threshold suppresses votes for low entropy valuesThreshold = Average BE

7、Frequency (Cohen et al)Most frequent set of tokens are assumed to be meaningful tokensFrequencies of tokens with length =1,2, 3., 6 Shorter tokens are inherently more frequent than longer toke nsNormalize frequencies for tokens of the same length using standard deviationBoundaries are assigned at th

8、e end of most frequent token in the windowrainy dayFrequency in window:(1 )T = 3(2),lf, = 5(3)=2(4)”lt is” = 3Mutual Information (Ml)Mutual Information given by:Gives us the reduction of uncertainty in presence of event b given event &Ml does not incorporate the counter evidence when & occurs withou

9、t b and vice versaAugmented Expected Mutual Information(AEMI)AEMI sums the supporti ng evide nee and subtracts the counter evidence For each window, the location with the minimum AEMI value suggests a boundaryItisRrainydayabMinimum Description Length(MDL)Shorter code assigned to frequent tokens to m

10、inimize the overall coding lengthBoundary yielding shortest coding length is assigned votesCoding Length per byte:Lg P(tj): no of bits to encode t(o |tj|=lengthoftjI疋渥淪京n藝IJE|rainy ay / t|efttjghtNormalize scores of each algorithm Each algorithm produces list of scores Since the number of votes is p

11、roportional to the score, the scores must be normalized Each score is replaced by the number of standard deviations that the score is away from the mean valueNormalize votes of each algorithmAlgorithms produce list of votes depending on the scoresMake sure each algorithm votes with the same weightNu

12、mber of votes is replaced by the number of standard deviations from the mean valueNormalizing Scores and VotesCombined Normalized VotesCombined Approach with WeightedVotingA list of votes from all the experts is gatheredFor each boundary, the final votes are summedA boundary is placed at a position

13、if the votes at the position exceed threshold. Threshold = Average number of VotesAnomaly Detection Algorithm 一 LERAD(Mahoney and Chan)LERAD forms rules based on 23 attributes o First 15 attributes: from packet headerNext 8 attributes: from the payloado Example Rule:If port = 80 then wordl 二 “GET”Or

14、iginal Payload attributes: space separated tokensOur Payload attributes: Boundary separated tokensExperimental Data2019 DARPA Intrusion Detection Evaluation Data Set Week 3 :attack free (training) dataWeeks 4, 5: attack containing (test) dataEvaluations A, B, C (Known boundaries) : Week 3o trained:

15、days 1 - 4o tested: days 5-7Prevent gaining knowledge from Weeks 4 and 5Evaluation D (Detected attacks)Trained: Week 3Tested :Weeks 4 and 5Evaluation A: % of Space-SeparatedTokens RecoveredMethodPort#25Port#80Port#21Port#79AvgFreq+MDL5226218145.0Freque ncy1516139936.0BE +AEMI + MDL+ Freq211451213.0A

16、EMI5943212.5MDL6732510.3BE33194.0Evaluation B: % of Keywords in RFCsRecoveredMethodPort#25Port#80Port#21AvgFreq+MDL40365945.0Freque ncy31284033.0BE+AEMI+MDL+Freq12132115.3AEMI9525.3MDL7614.7BE3222.3Evaluation C: Entropy of Output (Lower is Better) average across 6 portsMethodAverage ValueFreque ncy5

17、.0MDL5.03Freq+MDL5.06BE5.25BE +AEMI + Freq + MDL5.56AEMI6.38Ranking of AlgorithmsMethodEvaluation AEvaluation BEvaluation CFreq+MDL3Freque ncy221BE+AEMI+ MDL+ Freq335AEMI446MDL552BE664Detection Rate for Space Separated VsBoundary Separated (Freq + MDL)Port #10 FP/day100 FP/daySpaceBoundarySpaceBound

18、ary20224521141614172233332313141314251516161679333380101011131132222Overall59626368% Improvement58Summary of ContributionsUsed payload in formation, while most IDS concentrate on header information.Proposed AEMI + MDL for boundary detection Combined all and subset of algorithmsUsed weighted voting to indicate confidence Proposed techniques find boundaries better than spacesAchieved higher detection rates in an anomaly detection systemFuture WorkFurther evaluation on other portsPick more useful toke ns in stead of first 8DARPA data set is partially synthetic, further evaluation on real traffi

溫馨提示

  • 1. 本站所有資源如無特殊說明,都需要本地電腦安裝OFFICE2007和PDF閱讀器。圖紙軟件為CAD,CAXA,PROE,UG,SolidWorks等.壓縮文件請下載最新的WinRAR軟件解壓。
  • 2. 本站的文檔不包含任何第三方提供的附件圖紙等,如果需要附件,請聯系上傳者。文件的所有權益歸上傳用戶所有。
  • 3. 本站RAR壓縮包中若帶圖紙,網頁內容里面會有圖紙預覽,若沒有圖紙預覽就沒有圖紙。
  • 4. 未經權益所有人同意不得將文件中的內容挪作商業或盈利用途。
  • 5. 人人文庫網僅提供信息存儲空間,僅對用戶上傳內容的表現方式做保護處理,對用戶上傳分享的文檔內容本身不做任何修改或編輯,并不能對任何下載內容負責。
  • 6. 下載文件中如有侵權或不適當內容,請與我們聯系,我們立即糾正。
  • 7. 本站不保證下載資源的準確性、安全性和完整性, 同時也不承擔用戶因使用這些下載資源對自己和他人造成任何形式的傷害或損失。

評論

0/150

提交評論