論文對(duì)主要的網(wǎng)絡(luò)流量分類技術(shù)進(jìn)行了闡述,并提出了目前網(wǎng)絡(luò)流量分類技術(shù)所面臨的問(wèn)題。然后論文主要研究了以下內(nèi)容:準(zhǔn)確的即時(shí)通信(IM)流量分類方法,有效的特征選擇方法和用于IM流量分類的有效特征包數(shù)的界定方法。為了提高IM流量分類精度,穩(wěn)定性和分類性能,論文提出了一些算法模型,具體貢獻(xiàn)如下:1、論文在研究和分析即時(shí)通信流量的特點(diǎn)和相關(guān)機(jī)器學(xué)習(xí)的理論和方法的基礎(chǔ)上,首先選擇支持向量機(jī)(SVM)、C4.5決策樹(shù)、貝葉斯網(wǎng)絡(luò)和樸素貝葉斯這4種經(jīng)典機(jī)器學(xué)習(xí)分類器進(jìn)行即時(shí)通信文本流量分類的研究,并在兩種不同網(wǎng)絡(luò)環(huán)境下采集即時(shí)通信流量數(shù)據(jù)做為數(shù)據(jù)集。然后,提取了50個(gè)流量特征用于訓(xùn)練和測(cè)試。實(shí)驗(yàn)結(jié)果表明,所有分類器在準(zhǔn)確率、召回率和精度指標(biāo)方面都非常有效,但其中C4.5機(jī)器學(xué)習(xí)分類器的性能最佳。2、基于機(jī)器學(xué)習(xí)的流量分類中,不恰當(dāng)?shù)奶卣鬟x取容易產(chǎn)生錯(cuò)誤的流量分類結(jié)果,因此即時(shí)通信流量分類特征的選取也是即時(shí)通信流量分類中的一個(gè)挑戰(zhàn)性問(wèn)題。為了解決這個(gè)問(wèn)題,論文提出了一種特征選擇度量標(biāo)準(zhǔn)Weighted Mutual Information(WMI),在此基礎(chǔ)上提出了一個(gè)WMI_AC...

【文章頁(yè)數(shù)】：153 頁(yè)

【學(xué)位級(jí)別】：博士

【文章目錄】：
Abstract
摘要
Chapter1 Introduction
    1.1 Research Background
        1.1.1 Port based traffic classification
        1.1.2 Deep packet inspection
        1.1.3 Machine learning based traffic identification
    1.2 Survey of related work
        1.2.1 Network traffic classification feature description and extraction
        1.2.2 Supervised learning traffic classification
        1.2.3 Un-Supervised learning traffic classification
        1.2.4 Early stage traffic classification
    1.3 Practical Background of Network Traffic Classification
        1.3.1 Bayes net machine learning classifier
        1.3.2 Na?ve Bayes machine learning classifier
        1.3.3 Support vector machine learning classifier
        1.3.4 C4.5 decision tree machine learning classifier
    1.4 Internet trace traffic data set
    1.5 IM traffic classification result analysis
        1.5.1 Performance measurement
        1.5.2 Results and analysis
        1.5.3 Contributions for IM traffic classification research
    1.6 Structure of Thesis
Chapter2 Effective Feature Selection for IM Application Traffic Classification
    2.1 Introduction
    2.2 Feature Selection Metrics
        2.2.1 Mutual Information Based Metric
    2.3 Proposed Method
        2.3.1 Weighted Mutual Information(WMI)Metric
        2.3.2 ACC Metric
        2.3.3 WMI_ACC Algorithm
        2.3.4 Statistical Test
    2.4 Evaluation Methodology
        2.4.1 Data Sets
        2.4.2 HIT Trace I Dataset
        2.4.3 NIMS Dataset
        2.4.4 Performance Measures
    2.5 Experimental Results and Analysis
        2.5.1 Wilcoxon Pairwise Statistical Test Result
        2.5.2 Selected Features of Our Propose Algorithm
        2.5.3 Comparison
    2.6 Analysis and Discussion
    2.7 Summary
Chapter3 Feature Selection for Imbalance IM Applications Traffic Classification
    3.1 Introduction
    3.2 Related Work
    3.3 Methodology
        3.3.1 Feature Selection Metrics
        3.3.2 AUC Metric
        3.3.3 Feature Selection Algorithms
        3.3.4 WMI_AUC Algorithm
        3.3.5 RFS Algorithm
    3.4 Evaluation Methodology
        3.4.1 Data Sets
        3.4.2 Evaluation Criteria for Performance Measurements
    3.5 Experimental Results and Analysis
        3.5.1 Analysis Results of HIT Trace1 Dataset
        3.5.2 Analysis Results of NIMS Dataset
    3.6 Analysis and Comparison
    3.7 Summary
Chapter4 Robust Feature Selection Approach for IM Applications Traffic
    4.1 Introduction
    4.2 Methodology
        4.2.1 FSA and FEA Proposed Methods
        4.2.2 Feature Selection Based Metrics
        4.2.3 Symmetrical Uncertainty Based Metric:
    4.3 The Feature Selection Approach(FSA):
        4.3.1 The Proposed Algorithm
        4.3.2 The Feature Evaluation Approach(FEA):
    4.4 Evaluation Methodology
        4.4.1 Datasets
        4.4.2 Performance Measurement
    4.5 Experimental Results and Analysis
    4.6 Analysis and Comparison
    4.7 Summary
Chapter5 Effective Feature Packet for IM Application At Early Stage Traffic
    5.1 Introduction
    5.2 Data Sets
        5.2.1 HIT Lab Trace Dataset
        5.2.2 HIT Dorm Trace Dataset
    5.3 Proposed Model
    5.4 Methodology
        5.4.1 Machine Learning Classifiers
        5.4.2 Statistical Test
    5.5 Evaluation Criteria for Performance Measurement
    5.6 Results and analysis
    5.7 Mutual Information Results of the HIT Trace I Data Set
        5.7.1 Mutual Information Results of the HIT Trace II Data Set
        5.7.2 Analysis Results of HIT Lab Trace Data Set of the Text Messages
    5.8 Summary
Conclusions
References
詳細(xì)總結(jié) 基于機(jī)器學(xué)習(xí)的即時(shí)通信流量分類技術(shù)
Published Papers
Acknowledgement
Resume



本文編號(hào)：3945669