目前的入侵检测系统普遍使用基于特征库的方法进行检测,这种方法需要已知类别标签的数据集来进行训练,而得到这种数据集的代价一般都很大。因此,我们对“非监督学习”算法应用于异常检测的检测效果进行了评估,我们共评估了三种算法-聚类算法、K-近邻算法和一类SVM算法,其结果令人满意。 Current intrusion detection systems commonly use feature-based methods for detection. This method requires the training of data sets of known class labels, and the cost of obtaining such data sets is generally large. Therefore, we evaluated the detection effect of unsupervised learning algorithm applied to anomaly detection. We evaluated three algorithms-clustering algorithm, K-nearest neighbor algorithm and a class of SVM algorithms, with satisfactory results.