Outliers refer to data that is far away from conventional data objects, resulting from different mechanisms, and there is a clear gap from conventional data. Outlier detection algorithms include statistics-based, distance-based, density-based, and cluster-based algorithms. Spectral clustering has become a research hotspot in the field of clustering because it can cluster on arbitrary-shaped data samples. Through the verification of artificial data and real data sets, the experiments show that the outlier detection algorithm based on spectral clustering is superior to the traditional outlier detection algorithm. Spectral clustering algorithm is a clustering method based on structural mapping. In a data set containing outliers, spectral clustering can be used to describe the structural characteristics of the data set and detect the presence of outliers. The essence of spectral clustering is to convert the clustering problem into the optimal partitioning problem of graphs. The graph partitioning criteria include mincut, RatioCut, Ncut etc. The spectral clustering algorithm is generally divided into two steps. Firstly, a relationship graph is established based on the similarity measure of data sample points.Secondly, an algorithm is constructed to perform cluster segmentation on the graph. This paper has improved in two aspects: on the one hand, the Self-Tuning algorithm proposed by Zelnik-Manor and Perona is used to recalculate the scale parameters, and on the other hand, the k-means algorithm is optimized. The algorithm is compared with traditional outlier detection algorithm. The effectiveness analysis and results show that the algorithm proposed in this paper has more advantages in detecting outliers.

Tingfeng Liu, Weifeng Zhu. An Outlier Detection Algorithm Based on Spectral Clustering. Academic Journal of Computing & Information Science (2023), Vol. 6, Issue 13: 82-91. https://doi.org/10.25236/AJCIS.2023.061313.

[1] Huang Qiang, Ye Qing, Nie Bin, et al. Research on Outlier Identification Methods [J]. Software Guide, 2019 (6): 35-41.

[2] Zelnik Manor L, Perona P. Self-connecting spectral clustering [C] Advancements in Neural Information Processing Systems (NIPS) Cambridge, MA: MIT Press, 2004.

[3] Liu Xiangxiang, Zhang Linlin, Zhu Siqiao, et al. A Multi form Flexible Resource Aggregation Model for Virtual Power Plants Based on Grey Target Theory and Spectral Clustering [J]. China Electric Power, 2023, 56 (11): 104-112.

[4] Jean Claude C. From spectral cluster to uniform solvent estimates on compact manifolds [J]. Journal of Functional Analysis, 2024286 (2).

[5] Kunshuo X, Ge T, Boyu L, et al. Sub Classification and Type Identification of Weathering Silicate Glasses Based on Spectral Clustering [J]. Journal of Physics: Conference Series, 20232607 (1).

[6] Hongyuan Z, Xuelong L. Discretize Relaxed Solution of Spectral Clustering via a Non European Algorithm [J]. IEEE transactions on neural networks and learning systems, 2023.

[7] Xiao G, Hai Z, Xiangyu C. On the efficiency of higher order spectral clustering under weighted stochastic block models [J]. Computational Statistics and Data Analysis, 2024190.

[8] Li Fangshuo, Liu Lina, Cheng Zhijiong, et al. Low voltage household transformer relationship recognition based on improved spectral clustering algorithm [J]. China Test, 2023, 49 (10): 128-134.

[9] Ma Zhaohui, Hou Jichao, Xie Chengxin, et al. Research on Spectral Clustering Algorithm Based on User Characteristics [J]. Information Technology and Informatization, 2023, (10): 104-108.

[10] Guo Yingya, Wang Lijuan, Geng Haijun. Edge Server Placement Algorithm Based on Spectral Clustering [J]. Computer Science, 2023, 50 (10): 248-257.

[11] Jiayi T, Yuping L, Xinwang L. Multiview Spectral Clustering Based on Consensus Neighbor Strategy [J]. IEEE transactions on neural networks and learning systems, 2023, PP.