Presentation is loading. Please wait.

Presentation is loading. Please wait.

Intelligent Database Systems Lab Presenter: WU, JHEN-WEI Authors: Olatz Arbelaitz, Ibai Gurrutxaga, Javier Muguerza, Jesús M. Pérez and Iñigo Perona 2013.

Similar presentations


Presentation on theme: "Intelligent Database Systems Lab Presenter: WU, JHEN-WEI Authors: Olatz Arbelaitz, Ibai Gurrutxaga, Javier Muguerza, Jesús M. Pérez and Iñigo Perona 2013."— Presentation transcript:

1 Intelligent Database Systems Lab Presenter: WU, JHEN-WEI Authors: Olatz Arbelaitz, Ibai Gurrutxaga, Javier Muguerza, Jesús M. Pérez and Iñigo Perona 2013. PR. An extensive comparative study of cluster validity indices

2 Intelligent Database Systems Lab Outlines Motivation Objectives Cluster validity indices Experimental setup Results Conclusions Comments 2

3 Intelligent Database Systems Lab Motivation Many indices have been proposed, there is no recent extensive comparative study of their performance. 3

4 Intelligent Database Systems Lab Objectives To compare 30 cluster validity indices in many different environments with different characteristics. To build a guideline for selecting the most suitable index for each possible application. 4

5 Intelligent Database Systems Lab Cluster validity indices Dunn index (D ↑) 5 Calinski-Harabasz (CH ↑)

6 Intelligent Database Systems Lab Cluster validity indices Gamma index (G ↓) 6 C-Index (CI ↓)

7 Intelligent Database Systems Lab Cluster validity indices 7 Davies-Bouldin index (DB ↓) Silhouette index (Sil ↑)

8 Intelligent Database Systems Lab Cluster validity indices 8 Graph theory based Dunn and Davies-Bouldin variations (D MST ↑, D RNG ↑, D GG ↑, DB MST ↓, DB RNG ↓, DB GG ↓) Generalized Dunn indices (gD31 ↑, gD41 ↑, gD51 ↑, gD33 ↑, gD43 ↑, gD53 ↑)

9 Intelligent Database Systems Lab Cluster validity indices 9 S_Dbw index (SDbw ↓) CS index (CS ↓) Davies-Bouldin* (DB* ↓)

10 Intelligent Database Systems Lab Cluster validity indices 10 Sym-index (Sym ↑) Score function (SF ↑)

11 Intelligent Database Systems Lab Cluster validity indices 11 Point Symmetry-Distance (SymDB ↓, SymD ↑, Sym33 ↑) SymDB SymD Sym33

12 Intelligent Database Systems Lab 12 Cluster validity indices COP index (COP ↓) Negentropy increment (NI ↓)

13 Intelligent Database Systems Lab 13 Cluster validity indices SV-Index (SV ↑) OS-Index (OS ↑)

14 Intelligent Database Systems Lab Experimental setup 14

15 Intelligent Database Systems Lab Results 15

16 Intelligent Database Systems Lab 16 Results - Synthetic datasets

17 Intelligent Database Systems Lab Results – Real datasets 17

18 Intelligent Database Systems Lab 18 Results – Statistical tests

19 Intelligent Database Systems Lab Conclusions Some SVIs appear to be more suitable for certain configurations. The overall trend never changed dramatically when we focused on a particular factor. The results for real and synthetic datasets are qualitatively similar. 19

20 Intelligent Database Systems Lab Comments Contributions – Present the results of the most extensive CVI comparison ever carried out. – This comparison is the first extensive CVI comparison with the methodological correction. Applications – Build a guideline for selecting the most suitable index for each possible application. 20


Download ppt "Intelligent Database Systems Lab Presenter: WU, JHEN-WEI Authors: Olatz Arbelaitz, Ibai Gurrutxaga, Javier Muguerza, Jesús M. Pérez and Iñigo Perona 2013."

Similar presentations


Ads by Google