计算机工程与应用
計算機工程與應用
계산궤공정여응용
COMPUTER ENGINEERING AND APPLICATIONS
2013年
17期
89-93
,共5页
网络异常%概率神经网络%朴素贝叶斯分类器%融合%异常分类
網絡異常%概率神經網絡%樸素貝葉斯分類器%融閤%異常分類
망락이상%개솔신경망락%박소패협사분류기%융합%이상분류
network anomaly%Probability Neural Network(PNN)%Naive Bayes Classifier(NBC)%fusion%anomaly classification
对网络异常进行分类有利于管理员更好地管理网络,然而单一的分类器存在对各类异常的分类效果不均衡,不够全面等问题。鉴于此在研究了常用于分类的概率神经网络(Probability Neural Network,PNN)算法和朴素贝叶斯分类器(Naive Bayes Classifier,NBC)算法的基础上提出了一种融合NBC与PNN的网络异常分类模型。该模型将PNN与NBC对各类网络异常的分类精度作为权值,通过计算得出未知流量所属各类别的概率,最大值为预测结果,通过KDD99数据集对该模型进行测试,实验结果表明,提出的新模型相对于仅使用PNN或者NBC的单分类器,其对各类异常的分类效果具有更好的均衡性和更高的分类精度。
對網絡異常進行分類有利于管理員更好地管理網絡,然而單一的分類器存在對各類異常的分類效果不均衡,不夠全麵等問題。鑒于此在研究瞭常用于分類的概率神經網絡(Probability Neural Network,PNN)算法和樸素貝葉斯分類器(Naive Bayes Classifier,NBC)算法的基礎上提齣瞭一種融閤NBC與PNN的網絡異常分類模型。該模型將PNN與NBC對各類網絡異常的分類精度作為權值,通過計算得齣未知流量所屬各類彆的概率,最大值為預測結果,通過KDD99數據集對該模型進行測試,實驗結果錶明,提齣的新模型相對于僅使用PNN或者NBC的單分類器,其對各類異常的分類效果具有更好的均衡性和更高的分類精度。
대망락이상진행분류유리우관리원경호지관리망락,연이단일적분류기존재대각류이상적분류효과불균형,불구전면등문제。감우차재연구료상용우분류적개솔신경망락(Probability Neural Network,PNN)산법화박소패협사분류기(Naive Bayes Classifier,NBC)산법적기출상제출료일충융합NBC여PNN적망락이상분류모형。해모형장PNN여NBC대각류망락이상적분류정도작위권치,통과계산득출미지류량소속각유별적개솔,최대치위예측결과,통과KDD99수거집대해모형진행측시,실험결과표명,제출적신모형상대우부사용PNN혹자NBC적단분류기,기대각류이상적분류효과구유경호적균형성화경고적분류정도。
Classifying the network anomalies will help the administrators manage the network better. However, the single classi-fier has the problem that the results of the classification of the various of anomalies are not balanced, not comprehensive and other issues. In consideration of these facts, based on the research of the PNN algorithm and the NBC algorithm which are the most frequently used in classification filed, it proposes a new model using the fusion of the two algorithms. This model uses the accu-racy of PNN and NBC that to classify the anomalies as weights, by calculating to obtain the probability that belongs to each cate-gory of the unknown flow, and the biggest probability will be choosed as the result. According to the verification of the KDD99 data set, experimental results show that the proposed model has the better classification rate and better balance than the simple classifier which through the PNN or NBC algorithm.