电讯技术
電訊技術
전신기술
TELECOMMUNICATIONS ENGINEERING
2013年
9期
1207-1212
,共6页
赵小欢%夏靖波%连向磊%李巧丽
趙小歡%夏靖波%連嚮磊%李巧麗
조소환%하정파%련향뢰%리교려
网络流%流量分类%相关特征选择%自适应增强算法%组合分类器
網絡流%流量分類%相關特徵選擇%自適應增彊算法%組閤分類器
망락류%류량분류%상관특정선택%자괄응증강산법%조합분류기
network traffic%traffic classification%correlation-based feature selection%adaptive boosting algo-rithm%ensemble classifier
针对单一分类方法在训练样本不足的情况下对于小样本网络流分类效果差的特点,通过自适应增强(Adaptive Boosting,AdaBoost)算法进行流量分类。算法首先使用CFS(Correlation-based Feature Se-lection)特征选择方法从大量网络流特征中提取出少量高效的分类特征,在此基础上,通过AdaBoost算法组合决策树、关联规则和贝叶斯等5种单一分类方法实现流量分类。实际网络流量数据测试表明,基于AdaBoost的组合分类方法的准确率在所选的几种算法中是最高的,其能够达到98.92%,且相对于单一的分类算法,组合流量分类方法对于小样本网络流的分类效果具有明显提升。
針對單一分類方法在訓練樣本不足的情況下對于小樣本網絡流分類效果差的特點,通過自適應增彊(Adaptive Boosting,AdaBoost)算法進行流量分類。算法首先使用CFS(Correlation-based Feature Se-lection)特徵選擇方法從大量網絡流特徵中提取齣少量高效的分類特徵,在此基礎上,通過AdaBoost算法組閤決策樹、關聯規則和貝葉斯等5種單一分類方法實現流量分類。實際網絡流量數據測試錶明,基于AdaBoost的組閤分類方法的準確率在所選的幾種算法中是最高的,其能夠達到98.92%,且相對于單一的分類算法,組閤流量分類方法對于小樣本網絡流的分類效果具有明顯提升。
침대단일분류방법재훈련양본불족적정황하대우소양본망락류분류효과차적특점,통과자괄응증강(Adaptive Boosting,AdaBoost)산법진행류량분류。산법수선사용CFS(Correlation-based Feature Se-lection)특정선택방법종대량망락류특정중제취출소량고효적분류특정,재차기출상,통과AdaBoost산법조합결책수、관련규칙화패협사등5충단일분류방법실현류량분류。실제망락류량수거측시표명,기우AdaBoost적조합분류방법적준학솔재소선적궤충산법중시최고적,기능구체도98.92%,차상대우단일적분류산법,조합류량분류방법대우소양본망락류적분류효과구유명현제승。
To cope with the poor performance of single classification algorithms on minority flows when the train dataset is deficient,the AdaBoost (Adaptive Boosting)algorithm is introduced to classify network traffic . On the basis of selecting few but effective classification features with CFS (Correlation-based Feature Selection)method from a variety of flow′s features,the AdaBoost algorithm is used to combine five single classification algorithms which belong to Decision Tree,Rules and Bayes respectively for the sake of traffic classification . The experi-ment over real network traffic shows that the AdaBoost algorithm has the highest precision up to 98.92% amongthe selected classification algorithms . Moreover,the AdaBoost algorithm achieves great improvement on the per-formance of minority flows′ classification compared with single classification algorithms .