基于瀑布型混合技术的异常检测算法

Anomaly Detection Algorithm Based on Waterfall Hybrid Technology

针对隔离森林(i Forest:isolation Forest)算法对局部异常点检测能力较低,LOF(Local Outlier Factor)算法检测时间较长的问题,提出了基于瀑布型混合技术的隔离森林算法i Forest-WHT(isolation Forest based on Waterfall Hybrid Technology)。该算法借鉴瀑布型混合技术思想,将隔离森林算法作为过滤器,以分割路径为阈值判断依据,将路径小于阈值的数据放入候选异常子集,继而使用考虑极值影响的改进的LOF算法对候选异常子集进一步精化,得到更加精确的异常点。实验结果证明,该算法能以较高的效率识别局部异常点,提高了算法的F1值,并且降低原LOF算法的误检率。

iFores （isolation Forest） algorithm has a low detection ability for local outlier detection, and the detection time of LOF （Local Outlier Factor） algorithm is longer, and a improved algorithm which can solve these problems named iForest-WHT （isolation Forest based on Waterfall Hybrid Technology） is proposed. Based on the idea of waterfall hybrid technology, the iForest algorithm is used as the filter, the split path is the threshold judgment method, the data with path less than the threshold is put into the candidate anomaly subset. Then the improved LOF algorithm considering the extreme value is used to refine the candidate anomaly subset to obtain more accurate anomaly subset. The experimental results show that the algorithm can identify the outliers at higher efficiency, improve the F1 value of the algorithm and reduce the false alarm rate of the original LOF algorithm.