Diese Präsentation wurde erfolgreich gemeldet.
Wir verwenden Ihre LinkedIn Profilangaben und Informationen zu Ihren Aktivitäten, um Anzeigen zu personalisieren und Ihnen relevantere Inhalte anzuzeigen. Sie können Ihre Anzeigeneinstellungen jederzeit ändern.

FUAT – A Fuzzy Clustering Analysis Tool

2.540 Aufrufe

Veröffentlicht am

For Full Paper:


Veröffentlicht in: Technologie, Business
  • Als Erste(r) kommentieren

FUAT – A Fuzzy Clustering Analysis Tool

  1. 1. A. Selman BOZKIR - Ebru Akçapınar Sezer Hacettepe University – Computer Eng. Dept
  2. 2. <ul><li>What is clustering and FCM? </li></ul><ul><li>Principle of Fuzzy Clustering </li></ul><ul><li>The difficulties in FCM </li></ul><ul><li>Proposed solution: FUAT </li></ul><ul><li>Details </li></ul><ul><li>Conclusion </li></ul>Perspective
  3. 3. Clustering <ul><li>Cluster analysis or clustering is the task of assigning a set of objects into groups (called clusters) so that the distances of objects in the same cluster (intra class) are less than the distances in different cluster s (inter class) . </li></ul>
  4. 4. Clustering (Schemas) <ul><li>Hard Clustering (ex:k-means) Soft Clustering (ex: EM,FCM) </li></ul>each data element belongs to exactly one cluster elements can belong to more than one cluster, and associated with each element is a set of membership levels .
  5. 5. Fuzzy c-means clustering <ul><li>Based on Zadeh’s fuzzy sets theory. </li></ul><ul><li>Invented by Bezdek, 1981 </li></ul><ul><li>A soft clustering method </li></ul><ul><li>C ombines the c-means approach with the handling of the fuzziness existing in the data </li></ul><ul><li>one of the most popular unsupervised c lustering algorithm, which is widely used in pattern recognition, image recognition, gene classification, etc [1] </li></ul>
  6. 6. FCM in Principle <ul><li>c as an input parameter </li></ul><ul><li>segments data into fuzzy clusters by providing typical prototypes for each of them </li></ul><ul><li>link between objects and cluster prototypes are expressed via a membership matrix </li></ul><ul><li>where u ij is the membership degree of x i in the cluster j, m is a real number denoting the fuzziness coefficient greater than 1, x i is the i th of d-dimensional data and c j is the cluster centroid of cluster j. Further, fuzzy segmentation is done with the optimization </li></ul>
  7. 7. Difficulties of Fuzzy c-means clustering <ul><li>as stated by [ 2 ], three major difficulties were drawn ; </li></ul><ul><li>(1) how to detect optimal number of clusters ? </li></ul><ul><li>(2) how to choos e the initial cluster centroids ? </li></ul><ul><li>(3) how to evaluate cluster results, characterized by large variations in cluster shape, cluster density, and the number of points in different clusters </li></ul>
  8. 8. Solution: FUAT <ul><li>to analyze, explore and visualize different aspects of obtained fuzzy clusters </li></ul><ul><li>convert black box of fuzzy clustering to transparent box </li></ul>
  9. 9. FUAT – General Overview <ul><li>FCM and EM based clustering </li></ul><ul><li>Automatic cluster count estimator for non domain-experts </li></ul><ul><li>Various interactive viewers for different insights </li></ul><ul><li>Zooming, filtering, saving is available for results </li></ul><ul><li>CSV file support </li></ul><ul><li>R connectivity package (StatConn’s R(D)COM), ZedGraph and Microsoft GLEE is employed during the development </li></ul><ul><li>Developed at C#.NET </li></ul>
  10. 10. FUAT <ul><li>General FCM Settings and Membership Table </li></ul>
  11. 11. FUAT <ul><li>Automatic cluster count detection is based on Bayesian Information Criteria (BIC) implemented in EM framework of Mclust package of R. </li></ul>
  12. 12. FUAT <ul><li>Cluster Population Distribution Viewer </li></ul>
  13. 13. FUAT <ul><li>Cluster Centroids Viewer </li></ul>
  14. 14. FUAT <ul><li>Cluster Membership Histogram Viewer </li></ul>
  15. 15. FUAT <ul><li>Points of Interest Viewer </li></ul>
  16. 16. FUAT <ul><li>Cluster Dependency Viewer </li></ul>
  17. 17. Conclusion <ul><li>FUAT is useful at gaining insight from cluster analysis. </li></ul><ul><li>Ability for cluster analysis seperately and integrated to overcome difficulties of FCM usage </li></ul><ul><li>Software R can be used in native applications to power third party ML,DM applications via suitable interfaces. </li></ul><ul><li>Some Examples of Practical Benefits: Useful at revealing the inner structure of imbalanced data sets Useful at detecting important and dominant attributes in datasets </li></ul>
  18. 18. References [1] Jingwei Liu, Meizhi Xu, Kernelized fuzzy attribute C-means clustering algorithm, Fuzzy Sets and Systems 159 (2008) 2428 – 2445 [2] Dae-Won Kim , Kwang H. Lee, Doheon Lee, A novel initialization scheme for the fuzzy c-means algorithm for color clustering, Pattern Recognition Letters 25 (2004) 227–237
  19. 19. <ul><li>Thanks for listening …. </li></ul><ul><li>Questions? </li></ul>