Please explain how to cluster the single variable data using minitab software. Please share if any algorithm is available for making similar groups from a set of data. Let the data is, A, B, C, D, E, F, G, H, I, J, K, L and their respective scores 1, 5, 8, 7, 5, 12, 15, 13, 18, 23, 34 and 4. Please tell me a simple procedure that make non-overlapping groups of objects from the given data.
Thanks in advance.
It actually doesn't matter if you have the single variable or multidimensional data. Probably the simplest procedure is k-means. In case of the 1D data, the distance between two elements X and Y is |X-Y| (absolute value). K-means is included in many software, for example, ELKI. If you want to include k-means procedures in your own program code, you can find k-means here:
http://rosettacode.org/wiki/K-means%2B%2B_clustering
This code is written for 2D clustering, you can simplify it or assign the 2nd coordinate (y) equal to 0. You can also run command-line clustering programs like Cluster 3.0 from your program:
http://bonsai.hgc.jp/~mdehoon/software/cluster/software.htm
There are also many clustering libraries.
The only problem of the k-means method is that you must know how many cluster you will have.
Naveen,
To implement k-means in Minitab, as Lev suggests, from Minitab's Stat tab, choose Stat --> Mulitvariate --> Cluster K means. I would recommend that you also consider clustering your data using some of the hierarchical clustering schemes and graphical displays that Minitab also has available in Stat --> Mulitvariate --> Cluster Observations. Make sure to ask Minitab to 'show dendogram.' You'll see that you will need to specify a linkage and distance method - you might start off by using Minitab's defaults. Which clustering method that you ultimately choose (i.e., what linkage and distance methods) to use depends on what assumptions you are willing to make about your data. For example, k-means clustering and other methods that use a Euclidean distance metric are appropriate for data from a mixture of gaussians.
I am not sure whether any clustering tool will enable clustering based on just one variable. I trick that might help is to add another artificial variable - with exactly the same value for all individuals. This will provide you clustering driven by only your single variable of interest, since the artificial variable is a constant for all individuals.
Thanks. The clustering of single variable using minitab can be possible by using a dummy variable with a constant (even a zero column) under Average linkage method (which I have tested). However, Single variable clustering may be investigated using its distributional properties, inflexion points, etc... If you have any idea in doing so, please share with me. I am also trying in that way by considering an empirical distribution and a bootstrap distribution. Please suggest me.
Does anyone know of any references that give a good argument for why you would cluster on one variable? My question is whether this is a legitimate/useful approach to take.
Do you think that it is genuine to add another artificial variable with the very same values than the one I am interested in using as the clustering variable? that would be, in the example by Naveen Kumar Boiroju , scores for variable (1) 1, 5, 8, 7, 5, 12, 15, 13, 18, 23, 34 and 4; and a second row for variable (2) 1, 5, 8, 7, 5, 12, 15, 13, 18, 23, 34 and 4.
This is different from the artificial row with zero values for all samples.
I am facing the same problem. Adding a dummy variable (1) to trick minitab does not work, since it recognise it and do not allow you to use it. Nevertheless, using the creating a secodn column with the same value off the first one would result in bias distances and similarity within the clusters.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software