Outlier detection algorithm based on natural local core density ratio outlier factor
To address the challenges of recognizing unevenly distributed outliers and parameter dependency in existing outlier detection algorithms,a novel outlier detection algorithm was proposed based on the natural core density ratio outlier factor(ND-OF).The natural neighbor search algorithm was employed to adaptively determine the neighborhood parameter and compute the local density of data objects and their nearest neighbors.Subsequently,the local density of each data object was compared with its natural neighbors.If the object's density was lower than the maximum density among its natural neighbors,the comparison object was updated to the neighbor with the highest density.This iterative process continued until the highest density among the comparison objects was less than or equal to the object's own density,at which point the comparison process terminated.The natural neighbors involved in this process constituted the extended natural neighbors of the original data object,and the object with the highest density was identified as the local natural core point.Finally,the ratio of the local natural core point's density to the data object's density,combined with the influence of neighborhood size on outlier degree,was used as the outlier factor to measure the degree of outliers.Experiments on artificial and real datasets showed that the method had a higher F1 score than the comparison algorithms and showed good performance in outlier detection.
NETL
NSTL
万方数据
