基于免疫克隆选择加权朴素贝叶斯分类器的网络入侵检测

发布时间：2018-07-24 10:58

【摘要】：随着大数据时代的来临,网络数据流量成倍数增长,传统的网络入侵检测会出现崩溃、误报警和失效等问题,已经无法满足现阶段的需要。数据挖掘技术在处理大量数据方面有着得天独厚的优势,其在网络入侵检测领域的影响力也越来越大。朴素贝叶斯分类器作为传统的数据挖掘算法,具有诸多优点,能够很好的满足网络入侵检测的需要。但是其也存在着准确度不足的问题,针对该问题本文做了以下研究工作。对比研究了检测对象不同和分类分析方法不同的网络入侵检测,得出混合检测对象和异常检测网络入侵检测更适合现在的网络环境。根据经典网络入侵检测框架和数据挖掘的技术特点,提出了本文的网络入侵检测框架。深入研究了朴素贝叶斯分类器,并阐述了应用该方法实现网络入侵检测的原理。针对基于朴素贝叶斯分类器网络入侵检测的缺点,即假设各个特征值之间相互独立,提出了基于免疫克隆选择算法加权朴素贝叶斯分类器的网络入侵检测系统。其核心思想是把克隆选择算法的抗体作为权值,用抗体对传统朴素贝叶斯分类器幂加权,通过幂权值增加各个属性之间的相关性,用克隆选择算法不断优化权值来提高检测的准确度。为了验证免疫克隆选择加权朴素贝叶斯分类器的网络入侵检测正确性,通过MATLAB对该方法做了实验验证。KDD数据集在网络入侵领域具有权威性,本文实验用该数据集作为训练集和测试集。分别用MATLAB编程实现了朴素贝叶斯分类器和免疫克隆选择加权朴素贝叶斯分类器的功能。实验结果表明:本文提出的方法提高了传统朴素贝叶斯网络入侵检测的准确度。
[Abstract]:With the advent of the big data era, the network data flow increases in multiple, the traditional network intrusion detection will appear the problems of collapse, false alarm and failure, and can not meet the needs of the current stage. Data mining technology has a unique advantage in dealing with a large number of data, and its influence in the field of network intrusion detection is growing. As a traditional data mining algorithm, naive Bayesian classifier has many advantages and can meet the needs of network intrusion detection. However, it also has the problem of insufficient accuracy. In view of this problem, this paper has done the following research work. The network intrusion detection with different detection objects and different classification analysis methods is compared. It is concluded that the hybrid detection object and the anomaly detection network intrusion detection are more suitable for the current network environment. According to the technical characteristics of classical network intrusion detection framework and data mining, this paper proposes a network intrusion detection framework. The naive Bayesian classifier is studied and the principle of network intrusion detection using this method is described. Aiming at the shortcoming of network intrusion detection based on naive Bayesian classifier, that is, assuming that each eigenvalue is independent of each other, a network intrusion detection system based on immune clone selection algorithm weighted naive Bayesian classifier is proposed. Its core idea is to take the antibody of clonal selection algorithm as the weight value, to use the antibody to weighting the traditional naive Bayes classifier, and to increase the correlation between each attribute by the power weight. The clonal selection algorithm is used to optimize the weights to improve the accuracy of detection. In order to verify the correctness of network intrusion detection based on immune clone selection weighted naive Bayesian classifier, the method is tested by MATLAB to verify the authority of .kDD data set in network intrusion field. In this paper, the data set is used as training set and test set. The functions of naive Bayes classifier and immune clone selection weighted naive Bayes classifier are realized by MATLAB programming. Experimental results show that the proposed method improves the accuracy of traditional naive Bayesian network intrusion detection.
【学位授予单位】：兰州理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.08

【相似文献】