Density Peak Anomaly Detection Method Based on Rknn

In order to improve the detection accuracy of anomaly detection algorithms and reduce the dependence on data prior information when dealing with local anomalies, anomaly clusters and complex distribution data sets, a density peak anomaly detection method Rknn-DP based on inverse K nearest neighbors is proposed. First of all, the algorithm improved the calculation of local density and relative distance in the density peak algorithm by Rknn, to make it more accurately describe the characteristics of the abnormal points. After that, select points with low local density and high relative distance by adaptive threshold, as rough set of abnormal points. Finally, the Rknn method is used to prune the rough selection set, eliminate the noise point, reduce the associated error effect, and adaptively get the final abnormal set. Compared with ABOD, LSCP, HBOS, IForest algorithms in real data sets and artificial data sets, the results show that the Rknn-DP, algorithm performs with higher detection and adaptability.