1 / 16

APSCAN: A parameter free algorithm for clustering

APSCAN: A parameter free algorithm for clustering. Presenter : Cheng- Hui Chen Author : Xiaoming Chen, Wanquan Liu, Huining Qiu , Jianhuang Lai PRL 2011. Outlines. Motivation Objectives Methodology Experiments Conclusions Comments. Motivation.

drake
Télécharger la présentation

APSCAN: A parameter free algorithm for clustering

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. APSCAN: A parameter free algorithm for clustering Presenter: Cheng-Hui Chen Author: Xiaoming Chen, Wanquan Liu, HuiningQiu, Jianhuang Lai PRL 2011

  2. Outlines • Motivation • Objectives • Methodology • Experiments • Conclusions • Comments

  3. Motivation • There are two distinct drawbacks for DBSCAN: • It has two parameters are difficult to be determined. • DBSCAN does not perform well to datasets with varying densities

  4. Objectives • This paper propose a novel parameter free clustering algorithm named as APSCAN • As a parameter free clustering method, APSCAN is different from AP and DBSCAN. It is not only suitable for a single density data like DBSCAN but also can be used to cluster density varying datasets.

  5. Methodology

  6. Methodology MinPts = 2 Noise Class ID DBSCAN algorithm

  7. Methodology Responsibility Availabilities Candidate k Candidate examplar k Supporting data point I’ Competing Candidate k’ r(i, k) a(i, k) Data point i Affinity propagation clustering algorithm

  8. APSCAN Normalized density list generation

  9. APSCAN • The Double-Density Based SCAN (DDBCAN)

  10. Synthesize the result by Label Update Rule • If all the points in Cjof Resultj are noise points in Resulti, we give a new class label for all points in cluster Cjin the updated clustering result. j • If p is in the cluster Cjin Resultjand p is in the cluster Ci in Resulti, then mark p with label Ci in the updated clustering result. • If p belongs to Cj in Resultj and p is a noise point in Resulti, but not all the points in Cj are noise points in Resulti , we give p a label as j j i i If p is a noise point both in Resulti and Resultj , it is labeled as a noise point in the updated clustering result.

  11. Experiments

  12. Experiments Dataset Three Dataset One Dataset Two

  13. Experiments Toy dataset

  14. Experiments

  15. Conclusions • In this paper can conclude that the proposed APS-CAN has the following three advantages: • It is a parameter free clustering method. • Itis suitable for clustering datasets with varying densities. • Itcan preserve the irregular structure of a dataset.

  16. Comments • Advantages • It has achieved satisfactory performance on clustering datasets with varying densities. • Applications • Clustering

More Related