The outlier detection is to select uncommon data from a data set, which can significai improve the quality of results for the data mining algorithms. A typical feature of the outliers is that they are always far away from a majority of data in the data set. In this paper, we present a graph-ba outlier detection algorithm named INOD, which makes use of this feature of the outlier. The DistMean-neighborhood is used to calculate the cumulative in-degree for each data. The data, wh cumulative in-degree is smaller than a threshold, is judged as an outlier candidate. A KNN-ba selection algorithm is used to determine the final outlier. Experimental results show that the INOD algorithm can improve the precision 80% higher and decrease the error rate 75% lower than classical LOF algorithm.
展开▼
