An Internal Validity Index Based on Density-Involved Distance

https://ieeexplore.ieee.org/document/8672850

Requirement

The source code is written by Matlab r2016a. Versions lower than Matlab r2012a have not been tested.

Simplest Demo

• RUN Ncut_test.m to determine the optimal partition from varied partitions (produced by Jianbo Shi's Normalized cuts).

  CVDD.m includes Algorithm 1: CVDD in our paper.

  Ncut_test.m as an example includes Algorithm 2: CVDD-OP in our paper.

  'OP_CA' (in Ncut_test.m) shows the comparison. [ Purity, CVDD, CVNN, WB, Silhouette, CH, DB, Dunn, S_Dbw, I ]

Parameters in CVDD

No need to tune.

Datasets used

File Datasets_all30 includes 10 non-spherical clusters, 10 spherical clusters and 10 classification datasets (real datasets) used in the experiments of our paper.

Issues, Questions, etc

Please report issues here on the github page or contact "hly4ml@gmail.com"

Citation

If you find my code useful please consider citing:

@article{Hu2019,
  title={An Internal Validity Index Based on Density-Involved Distance},
  author={Hu, Lianyu and Zhong, Caiming},
  journal={IEEE Access},
  year={2019},
  volume={7},
  pages={40038-40051},
  doi={10.1109/ACCESS.2019.2906949}
}