fuzzy c means

p107027

Can I do the cluster model visualizer for Fuzzy c means clustering in rapid miner?

marcin_blachnik

Hi,

Fuzzy C-means was developed before the Cluster Visualizer emerged, therefore it is not compatible with the cluster visualizer operator. Although you can obtain similar information or even more when you mark "Add partition matrix" which will generate additional attributes which hold information about the fuzzy membership function - the so-called fuzzy partition matrix (see the confidence attributes returned on "ori" output port). With it you can make a scatter plot where on the X-axis is a given input variable and on the Y-axis are attributes called confidence(cluster_0), etc. 
For example, below is a visualization of the importance of values of attribute a3 of the iris dataset for all three clusters

Or you can do a scatter plot of different input attributes and color it with the confidence of a given cluster

so that you can see how strong given samples belong to the particular cluster.
Or apply PCA to reduce the input future space to just two attributes and repeat the above visualization with the confidence of the given cluster.

I hope that was helpful
Best regards
M. Blachnik

p107027

Hi,
Thanx for you suggestion!

marcin_blachnik

Hi,

Fuzzy C-means was developed before the Cluster Visualizer emerged, therefore it is not compatible with the cluster visualizer operator. Although you can obtain similar information or even more when you mark "Add partition matrix" which will generate additional attributes which hold information about the fuzzy membership function - the so-called fuzzy partition matrix (see the confidence attributes returned on "ori" output port). With it you can make a scatter plot where on the X-axis is a given input variable and on the Y-axis are attributes called confidence(cluster_0), etc. 
For example, below is a visualization of the importance of values of attribute a3 of the iris dataset for all three clusters

Or you can do a scatter plot of different input attributes and color it with the confidence of a given cluster

so that you can see how strong given samples belong to the particular cluster.
Or apply PCA to reduce the input future space to just two attributes and repeat the above visualization with the confidence of the given cluster.

I hope that was helpful
Best regards
M. Blachnik

p107027

Hi,
Thanx for you suggestion!

andri_setiawan

May I ask for  fuzzy c-means extension