A parameter-free similarity graph for spectral clustering

被引:30
作者
Inkaya, Tulin [1 ]
机构
[1] Uludag Univ, Dept Ind Engn, TR-16059 Bursa, Turkey
关键词
Spectral clustering; Similarity graph; k-nearest neighbor; epsilon-neighborhood; Fully connected graph; CONSTRUCTION; DENSITY;
D O I
10.1016/j.eswa.2015.07.074
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Spectral clustering is a popular clustering method due to its simplicity and superior performance in the data sets with non-convex clusters. The method is based on the spectral analysis of a similarity graph. Previous studies show that clustering results are sensitive to the selection of the similarity graph and its parameter(s). In particular, when there are data sets with arbitrary shaped clusters and varying density, it is difficult to determine the proper similarity graph and its parameters without a priori information. To address this issue, we propose a parameter-free similarity graph, namely Density Adaptive Neighborhood (DAN). DAN combines distance, density and connectivity information, and it reflects the local characteristics. We test the performance of DAN with a comprehensive experimental study. We compare k-nearest neighbor (KNN), mutual KNN, epsilon-neighborhood, fully connected graph, minimum spanning tree, Gabriel graph, and DAN in terms of clustering accuracy. We also examine the robustness of DAN to the number of attributes and the transformations such as decimation and distortion. Our experimental study with various artificial and real data sets shows that DAN improves the spectral clustering results, and it is superior to the competing approaches. Moreover, it facilitates the application of spectral clustering to various domains without a priori information. (C) 2015 Elsevier Ltd. All rights reserved.
引用
收藏
页码:9489 / 9498
页数:10
相关论文
共 40 条
[21]   The latest research progress on spectral clustering [J].
Jia, Hongjie ;
Ding, Shifei ;
Xu, Xinzheng ;
Nie, Ru .
NEURAL COMPUTING & APPLICATIONS, 2014, 24 (7-8) :1477-1486
[22]   Latent tree models for rounding in spectral clustering [J].
Liu, April H. ;
Poon, Leonard K. M. ;
Liu, Teng-Fei ;
Zhang, Nevin L. .
NEUROCOMPUTING, 2014, 144 :448-462
[23]   Non-negative and sparse spectral clustering [J].
Lu, Hongtao ;
Fu, Zhenyong ;
Shu, Xin .
PATTERN RECOGNITION, 2014, 47 (01) :418-426
[24]   HOW THE RESULT OF GRAPH CLUSTERING METHODS DEPENDS ON THE CONSTRUCTION OF THE GRAPH [J].
Maier, Markus ;
Von Luxburg, Ulrike ;
Hein, Matthias .
ESAIM-PROBABILITY AND STATISTICS, 2013, 17 :370-418
[25]  
Nadler B., 2006, P INT C ADV NEUR INF, V19, P1017
[26]   Spectral methods for graph clustering - A survey [J].
Nascimento, Maria C. V. ;
de Carvalho, Andre C. P. L. F. .
EUROPEAN JOURNAL OF OPERATIONAL RESEARCH, 2011, 211 (02) :221-231
[27]  
Ng AY, 2002, ADV NEUR IN, V14, P849
[29]   Normalized cuts and image segmentation [J].
Shi, JB ;
Malik, J .
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, 2000, 22 (08) :888-905
[30]   Vector quantization based approximate spectral clustering of large datasets [J].
Tademir, Kadim .
PATTERN RECOGNITION, 2012, 45 (08) :3034-3044