本文中,我們研究了協(xié)同聚類,并將相關(guān)概念與信息安全中的聚類分析聯(lián)系起來;在這個(gè)問題中,我們關(guān)注于紛繁復(fù)雜的網(wǎng)絡(luò)攻擊時(shí)代中,隨著數(shù)據(jù)的數(shù)量和復(fù)雜性不斷增長(zhǎng)所帶來的數(shù)據(jù)安全與隱私問題。在應(yīng)用需求的推動(dòng)下,我們引入了協(xié)同聚類框架,該框架能夠?yàn)樾畔踩械臄?shù)據(jù)挖掘應(yīng)用進(jìn)行大型分布式數(shù)據(jù)庫(kù)建模和網(wǎng)絡(luò)建模。協(xié)同聚類符合信息安全中的數(shù)據(jù)挖掘需求主要體現(xiàn)在兩個(gè)方面:首先是協(xié)同聚類能通過使用信息顆粒保證隱私,同時(shí)允許使用原型進(jìn)行協(xié)同任務(wù);其次,在面向具有高維大數(shù)據(jù)集和表示被監(jiān)控對(duì)象行為的多個(gè)特征時(shí),為算法提供可擴(kuò)展性,這反過來不僅增加了學(xué)習(xí)正常行為問題的復(fù)雜性,而且還可能給聚類分析帶來嚴(yán)重錯(cuò)誤。然而,諸如協(xié)同模糊聚類、協(xié)同自組織映射和協(xié)同生成式拓補(bǔ)映射等協(xié)同聚類方法存在需要輸入?yún)?shù)來決定協(xié)同信息影響的問題,這些參數(shù)對(duì)聚類結(jié)果又很大的影響,因此不能被忽視。我們提出了一種協(xié)同聚類框架,該框架使用粒子群優(yōu)化來最小化聚類的熵,以尋找最佳聚類中心。此外,它使用粒子矢量位置更新來確定協(xié)同信息的重要性,從而消除了對(duì)用戶輸入?yún)?shù)的依賴。被稱為粒子子圈的框架結(jié)合了來自幾種聚類算法的信息,從而部分解決了選擇正確聚類方法的問...

【文章頁(yè)數(shù)】：102 頁(yè)

【學(xué)位級(jí)別】：碩士

【文章目錄】：
摘要
Abstract
LIST OF SYMBOLS
CHAPTER1:INTRODUCTION
    1.1 OVERVIEW
    1.2 CLUSTERING IN INFORMATION SECURITY
    1.3 CHALLENGES
    1.4 APPLICATIONS
    1.5 CONTRIBUTION OF THIS THESIS
CHAPTER2:LITERATURE REVIEW
    2.1 CHALLENGES IN TRADITIONAL CLUSTER ANALYSIS
    2.2 DISTANCE AND SIMILARITY MEASURES
        2.2.1 Distance measures
        2.2.2 Similarity measures
    2.3 PROTOTYPE-BASED CLUSTERING ALGORITHMS
        2.3.1 K-Means
        2.3.2 Fuzzy C-Means
        2.3.3 Gaussian Mixture Models
        2.3.4 Affinity Propagation Clustering
    2.4 CLUSTER VALIDITY INDEXES
        2.4.1 Internal Validity Indexes
        2.4.2 External Validity Indexes
    2.5 CHAPTER CONCLUSION
CHAPTER3:COLLABORATIVE CLUSTERING
    3.1 CHALLENGES AND MODERN CLUSTER ANALYSIS
    3.2 COLLABORATION SCHEMES
    3.3 STATE OF THE ART IN COLLABORATIVE CLUSTERING
        3.3.1 Collaborative Fuzzy c-means clustering
        3.3.2 Prototype based collaborative algorithms
    3.4 CHAPTER CONCLUSION
CHAPTER4:PARTICLE SUBSWARMS COLLABORATIVE CLUSTERING
    4.1 THE COLLABORATIVE FUZZY CLUSTERING AND ITS CHALLENGES
    4.2 DEFINITIONS
    4.3 THE FRAMEWORK FUNDAMENTALS
        4.3.1 Fitness Function
        4.3.2 Particle Position Update
        4.3.3 Stopping Criteria
    4.4 THE DESIGN AND COMPLEXITY ANALYSIS
        4.4.1 Collaboration with crisp clustering
        4.4.2 Collaboration with fuzzy clustering
    4.5 THE EXPERIMENTAL RESULTS
        4.5.1 Crisp clustering results
        4.5.2 Fuzzy clustering results
        4.5.3 Comparison with other frameworks
    4.6 CHAPTER CONCLUSION
CHAPTER5:CONCLUSION
    5.1 PRIMARY FINDINGS
    5.2 LIMITATIONS OF PSSCC
    5.3 RECOMMENDATIONS FOR FUTURE WORK
ACKNOWLEDGEMENTS
REFERENCE
APPENDIX A:DATA SETS AND IMPLEMENTATIONS
    A.1 EXPERIMENTAL DATA SETS
    A.2 IMPLEMENTATIONS
APPENDIX B:CONVERGENCE AND PROOFS
    B.1 CONVERGENCE OF K-MEANS
    B.2 CONVERGENCE OF FUZZY C-MEANS
        B.2.1 Expectation step
        B.2.2 Maximization step
    B.3 CONVERGENCE OF GAUSSIAN MIXTURE MODELS
        B.3.1 Updating the mixing coefficients
        B.3.2 Updating the centers of the clusters
        B.3.3 Updating the covariance matrices
        B.3.4 Corollary:EM algorithm and Gaussian Mixtures



本文編號(hào)：3792481