基于专家反馈的广义孤立森林异常检测算法OACSTPCD

Aiming at the problem that the isolation forest algorithm cannot detect local anomalies parallel to the axes and the tree structure is unable to be dynamically updated,this paper proposed a generalized isolation forest anomaly detection algo-rithm based on expert feedback.Firstly,it projected the data to the sampled normal unit vector,and selected a split point from the mapping area to divide the data space,then repeated these operations until constructed a generalized isolation tree.Second-ly,it introduced the weights of the leaf nodes of each tree in the generalized isolation forest,which comprehensively considered the influence of the number of subspace partitions and the sample size in the subspace on anomaly scores.Finally,it calculated the weighted anomaly scores of each data,and submitted data with high anomaly scores to expert for batch labeling,then the al-gorithm updated the weights of the leaf nodes according to the labeling results,so as to dynamically adjust the structure of the generalized isolation tree.The experimental results show that the numbers of real abnormal data are marked by expert in 7 data-sets are better than that of the other tree-based anomaly detection algorithms,and the average precision in 12 datasets are 38.952％,49.144％and 49.144％higher than isolation forest,extended isolation forest,generalized isolation forest,respectively.