Multi-stage Mixed Attribute Outlier Detection Algorithm Based on Neighborhood Density Difference

With the development of big data, an increasing number of mixed attribute dataset has become ubiquitous. It is also extremely important for decision analysis in the processing of a mixed attribute dataset. The existing outlier detection algorithm does not process a mixed attribute dataset and some special objects, because of too high computational time complexity and unsatisfactory detection results. In this paper, we present a multi-stage mixed attribute outlier detection algorithm. Firstly, with data set being divided, the neighborhood information was constructed based on the heterogeneous similarity metric to generate the core point. Then, the primitive clusters can be formed on the basis of the definition. Finally, a neighborhood density difference metric-based outlier detection algorithm was designed to construct neighborhood outlier factor (NOF). Extensive experimental results show the advantages of the proposed method, which could improve the outlier detection accuracy and reduce the time complexity on mixed attributed.