I have analyzed the grains of some barley lines for some nutritional factors. Now, I need to cluster them to have the best group for nutritional value. What is the best way of clustering? I have tried hierarchical clustering and K-means clustering and I am not sure if they are suitable for this purpose or not, because actually I want it to tell which group is better with respect to their nutritions.
Dear Zaynab,
You can use PAST software for cluster and PCA analysis. It is freely available. In this software you can do cluster as well PCA analysis.
Other than the heuristical hierarchical and K-means methods, there are also optimal clustering methods for which you optimize an objective subject to constraints. So, how do you define "the best group for nutritional value"? That is the main thing for determining which are close to each other....
David,
Thank you. The best one should be the one with high nutritional factors. I have ranked all lines in excel for every factor and then added all rankings related to each line. Ranked the total ranks at the end and the results is different . I am not sue if I look at this correctly. That is why I am asking. Because hierarchial clustering categorize them according to their distance from center of the group and what it gives me as distinct groups is different with my expectations.
This is actually not a clustering problem because you are not trying to find groups within these barley lines. Can you came up with a formula to score each barley strain according to some gold-standard barley? Alternatively, if you can find nutritional data similar to what you have about the good strains, you can add them to your data and then run some regression to find the best strains in your data that are close or even outperforms your gold-standard barleys.
Good luck!
K-means is a good clustering method. You can also try the Self Organizing Maps (SOM) for your work. SOM is relatively new, interesting and more efficient. The results will be repeatable whereas with K-means you may get varied clusters each time since the seed (or centroid) may be at random value each time.
I am sure that you might have stuck up with the "number of clusters" issue.
Whatever the clustering method may be, you must be sure as to how many clusters you need. You should be able to then label as what each cluster means to you (and the real world). Try with optimum number of clusters... not too many and not too less. This also means "granularity". More number of clusters will dilute your work whereas less number of clusters may give confusing groups.
Once after obtaining clusters, you may like to remove unwanted clusters and further perform clustering so that you will get the required set having highest nutritional values.
- Badges
- Science method
- Similar topics
- Methodology