A clustering algorithm organises items into groups based on a similarity criteria. The Fuzzy c-Means algorithm is a clustering algorithm where each item may belong to more than one group (hence the word fuzzy), where the degree of membership for each item is given by a probability distribution over the clusters.

In this project, we Implement the Fuzzy c-Means Algorithm. For further details, please view the CWEB generated documentationfcm.pdf.

There are two separate programs in this project. They are:

This is an implementation of the Fuzzy C-Means Algorithm. It expects an input data file in the following format.

<number of data-points>
<number of clusters>
<number of dimensions>
<fuzzyness coefficient>
<termination criterion>
<data points> ...

For example, the following is a valid input file:

10 3 2
2.0 0.00005
897 802 765 992 1 521 220 380 729 969
184 887 104 641 909 378 724 582 387 583 

This has 10 data-points, each with 2 dimensions. The FCM program will try to cluster these 10 points into 3 clusters. The fuzzyness coefficient is 2.0 and the termination criterion (accuracy of the clustering) is 0.00005. For each of the 10 points, there are two values for each of the two dimensions. Hence, there are 20 data points listed.

Since it is tedious to generate the data points by hand, the gen program generates the data points automatically. In order to use this program, provide a filename which will be used for storing the input data for running fcm later.

An example run of the above two programs will be (in sequence)

$ cd fcm
$ make
$ ./gen input.dat
$ ./fcm input.dat

If the number of dimensions is 2 (points on a plane), and if you have access to gnuplot, then the graph plotting the clusters, with different colours for each of the clusters, can be generated by processing the file gnuplot.script with gnuplot. An example run will be

$ gnuplot gnuplot.script

This will generate an image file named cluster_plot.png.