Multi-view outlier detection based on multi-granularity fusion of fuzzy rough granules

In recent years, multi-view data has seen widespread application across various fields, presenting both opportunities and challenges due to its complex distribution across different views. Detecting outliers in such heterogeneous data has become a significant research problem. Existing multi-view outlier detection methods often rely on clustering assumptions, pairwise constraints between views, and a focus on learning consensus information, which overlook the inherent differences across views. To address the aforementioned issues, this paper proposes an outlier detection method based on the fusion of multi-granularity fuzzy rough information (MGFMOD). The method calculates a multi-granularity similarity matrix using fuzzy similarity relationships, combines similarity matrices from different granularities to form an upper approximation matrix, and constructs fused upper approximation granules to detect attribute anomalies. Neighbor domain probabilistic mapping is then employed to unify neighborhood relationships across views, allowing the analysis of both consistency and distribution differences to capture class outliers. Additionally, this paper employs a novel coarse-to-fine approximation method to construct the upper approximation matrix, further improving the accuracy of attribute outlier detection. Experimental results on multiple public datasets demonstrate that the proposed method generally outperforms existing multi-view outlier detection methods in terms of detection accuracy and robustness.