* |
A new initialization method for Fuzzy C-Means Algorithm using Fuzzy Subtractive Clustering (fzSC) |
|
We present a novel algorithm to address the
problem of parameter initialization of the standard FCM
algorithm. We have proposed a new subtractive clustering
method that uses fuzzy partition of the data instead of the
data themselves. The advantages of fzSC are that, unlike
traditional SC methods, it does not require specification of
the mountain peak and mountain radii, and, with a running
time of O(cn) compared to O(n^2) for the traditional SC
method, it is more efficient for large datasets. In addition,
our method can be integrated easily with fuzzy clustering
algorithms to search for the best centers of cluster
candidates.
|
|
|
Presentation
|
This will be presented at the WORLDCOMP 2011 conference, the World Congress in Computer Science, Computer Engineering,
and Applied Computing (July 18-21, 2011, Las Vegas, Nervada USA),
scheduled on July 19th 2011 at 10:00 - 10:20am.
|
|
Online demo
|
This online demo was developed using flash applet for client side and PHP scripting language with the fzsc program in C++ as mentioned above for server side.
|
|
Program source code
|
This algorithm was implemented in C++ and supported on Linux and MS Windows. This sourcecode can be compiled with either
the cl compiler for MS-Windows or the make macro on Linux.
windows-based: cl /EHsc fzsc.cpp data.c cluster.c command.c
linux-based: make
To run the program:
fzsc
This simple command will list all the parameters accepted by the program.
|
|
|
The table below shows the performance of our proposed method (fzSC) and some popular cluster indices,
including Bezdek partition coefficient (PC) and partition entropy (PE), on the 200 artificial datasets with
the number of clusters from 3 to 9.
fzSC and PBMF outperformed other cluter indices.
|
|
#c | fzSC | CF | CWB | FS | PBMF | BR | XB | PC | PE |
3 | 1.00 | 0.00 | 1.00 | 0.42 | 1.00 | 0.83 | 0.42 | 0.42 | 0.42 |
4 | 1.00 | 0.00 | 1.00 | 0.92 | 1.00 | 1.00 | 0.83 | 0.92 | 0.92 |
5 | 1.00 | 0.00 | 0.83 | 0.83 | 1.00 | 1.00 | 0.75 | 0.75 | 0.75 |
6 | 1.00 | 0.00 | 0.58 | 0.92 | 1.00 | 0.92 | 0.58 | 0.92 | 0.83 |
7 | 1.00 | 0.00 | 0.58 | 0.83 | 1.00 | 0.67 | 0.67 | 0.83 | 0.83 |
8 | 1.00 | 0.00 | 0.67 | 1.00 | 1.00 | 0.83 | 0.92 | 1.00 | 0.92 |
9 | 1.00 | 0.00 | 0.33 | 0.92 | 1.00 | 0.83 | 0.67 | 0.92 | 0.67 |
|
|
|
Artificial datasets
|
This packet contains 200 artificial datasets we used in testing this algorithm.
The results are shown in the figures 1,2 and the tables 1,2 of the manuscript.
|
|
|
Real datasets
|
This packet contains Iris, Wine, Glass and Breast Cancer Wisconsin datasets.
The results are shown in tables 3 of the manuscript.
|
|