Abstract
Anomaly detection is an important research direction in the ๏ฌeld of data mining and industrial dataset preprocess. The paper proposed a kernel neighbor densityde๏ฌnition with parallel computing mechanism for anomaly detection algorithm. The kernel neighbor density formula calculates the density of points in high dimensional space. In our de๏ฌnition, we adopt the median operation because the breakdown point of the median is the largest possible. So thede๏ฌnition could be a very robust estimate of the data location, and parallel computing mechanism is introduced to improve the ef๏ฌciency of algorithms. We use two real datasets and three different kernel functions to evaluate the performance of algorithms. The experiment results con๏ฌrm that the presentedde๏ฌnition of kernel neighbor density improves the performance of algorithms and the Gaussian kernel function has the best effect
