An outlier mining algorithm based on local weighted k-density

Most local outlier mining algorithms are inefficient and depend on many pre-set parameters when applied to high-dimensional dataset. In the paper, an outlier mining algorithm based on local weighted k-density is presented. Firstly, the attribute abnormal degree of each object in neighborhood for each attribute is computed by using local attribute entropy. Secondly, the corresponding attribute weight vector is set automatically according to attribute abnormal degree, so that man-made factor is reduced. Thirdly, the time-consuming and unnecessary step of re-calculating neighborhood is removed for simplifying computation, then local weighted outlier factor in the same neighborhood as getting attribute weight is calculated and outliers are detected based on local weighted k-density. In the end, the experimental results validate that the algorithm is feasible and efficient for high-dimensional outlier mining by utilizing star spectrum data.