基于最大局部密度间隔的特征选择方法

摘要

针对虚拟机数据特点及特征筛选问题, 借鉴局部异常因子算法中的"局部"思想, 提出基于最大局部密度间隔的特征评估准则, 通过最大化正常数据和异常数据的局部密度差异选出有效的特征子集;结合顺序后退搜索策略与提出的特征评估准则设计相应的特征选择算法, 筛选出有利于分类的虚拟机特征.实验结果表明, 所设计的特征选择算法能够有效处理虚拟机的类不平衡数据, 筛选出重要的虚拟机数据特征, 使数据的检测率和可理解性得到有效提升, 相比现有算法具有更好分类效果与更强适用性, 在相同条件下具有更小的计算开销.%For the characteristics of virtual machine data and the problem of its feature selection, using the local method of local outlier factor, the feature evaluation criterion based on maximum local density margin was proposed.The effective feature subsets were picked out by maximizing the density between normal and abnormal data.Combining the proposed criterion with sequential backward search algorithm, the corresponding feature selection algorithm was designed that might single out the virtual machine features which were beneficial to classification.The test results show that the proposed feature selection algorithm can effectively deal with the class-imbalanced data and single out the important features of virtual machine, with improvement of detection rate and understandability of virtual machine data.Compared with existing algorithms of feature selection, the proposed algorithm has better classification effects and stronger applicability with less computational cost under the same conditions.