.png){kind=icon}
.png){kind=icon}
Return
Multi-Kernelized Fuzzy Granular Outlier Detector
.png){kind=icon}
PRE
AI
Z
S
J
X
T
郑
DOI：10.1109/tkde.2026.3708765.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Fuzzy information granulation theory offers an effective framework for identifying uncertain outliers. However, existing kernelized fuzzy outlier detectors predominantly rely on a single kernel function, overlooking the potential of integrating diverse kernel information, which may lead to biases in outlier detection. Multi-kernel information fusion, by leveraging multiple kernel functions to extract information and employing information fusion strategies, can more comprehensively capture the characteristics of data distributions. Building on this, this study proposes a Multi-Kernelized Fuzzy Granular Outlier Detector (MKFGOD). The proposed detector first employs multiple kernel functions to extract fuzzy relation matrices from the data. It then integrates the multi-kernel information using various fusion strategies. Based on the fused multi-kernelized fuzzy relation matrix, the detector defines multi-kernelized fuzzy k-nearest neighbors and subsequently calculates the multi-kernelized fuzzy neighborhood density. Finally, the outlier degree of each sample is quantified by calculating the deviation in the multi-kernelized fuzzy neighborhood density. Experimental validation on multiple public datasets demonstrates that the proposed method significantly outperforms existing single-kernel and mainstream outlier detectors in terms of detection performance. This study not only enhances the effectiveness of outlier detection but also provides novel theoretical and practical guidance for the application of multi-kernel information fusion in fuzzy information granulation. Furthermore, it underscores the critical role of information fusion in complex data analysis.
Keywords:
Fuzzy information granulation theory
multi-kernel information fusion
kernelized fuzzy relations
fuzzy k-nearest neighbor
outlier detection
Journal
IF:
10.4
Papers:
6.7K
Citations:
3.2W
Organization
Cited Papers