改进的随机森林分类器网络入侵检测方法

Improved random forest classifier network intrusion detection method

目前网络入侵检测方法大多基于改进的机器学习算法,但是机器学习算法会出现过拟合情况,导致入侵检测准确率降低。为解决该问题,提出一种改进的随机森林分类器网络入侵检测方法,通过高斯混合模型聚类算法将数据分成不同的簇,为每一个簇训练不同的随机森林分类器,通过这些训练好的随机森林分类器进行网络入侵检测。训练和实验数据采用NSL-KDD网络入侵数据集,实施中首先根据属性比率数据特征提取方法进行数据处理,然后进行高斯混合聚类,最后使用随机森林分类器对聚类结果进行训练。实验结果表明,该方法相比其它机器学习算法具有更高的入侵检测准确率。

At present,network intrusion detection methods are mostly based on improved machine learning algorithms,but neural network algorithms sometimes have an over-fitting situation,which may lead to low accuracy of intrusion detection.To solve this problem,an improved random forest classifier network intrusion detection method was proposed.Dividing data into different clusters using Gaussian mixture model clustering algorithm and different random forest classifiers were trained for each cluster.The network intrusion detection was performed through these trained random forest classifiers.NSL-KDD network intrusion dataset was used as experimental simulation.Data processing was performed based on feature selection using attribute ratio,then Gaussian mixture clustering was performed,and finally the clustering results were trained using a random forest classifier.Experimental results show that the proposed method has higher intrusion detection accuracy compared with the other machine lear- ning algorithms.