Multi-View Outlier Detection Based on High-Order Neighbor Information

In recent years, the analysis and mining of multi-view data has become a new research hot spot. As an important branch, multi-view outlier detection has achieved some success, but there are still some problems to be solved. Most existing outlier detection methods assume that the instances conform to a clear clustering structure, while outliers exhibit unusual clustering characteristics in multiple views, and often apply some off-the-shelf clustering methods to discover the clustering structure. However, in real world applications, the structure of data is often complicated and the data may not reveal such a clear clustering structure. These methods are difficult to handle such complicated data. To address this issue, in this paper, we propose a simple yet effective method, named the multi-view outlier detection based on High-order Neighbor Information (HoNI), which makes full use of the high-order neighbor information of the data to better detect outliers. With the help of the high-order neighbor information, we do not need any assumptions of the clustering structure and thus do not need to apply any clustering methods. We conduct extensive experiments to verify the effectiveness and efficiency of the proposed method in terms of performance and running time on some benchmark data sets. The experimental results demonstrate the superiority of the proposed method.