For example, let me suppose that I have 100 thousands of geospatial data in my dataset, and I want to extract certain group that which is the most crucial. So I decided to do clustering and pick 'one cluster that seems to be most appropriate for my purpose of analysis' in the visualized result. Let me call this 'target cluster'. What I want ultimately is, mapping every single data in the target cluster on the map and doing spatial analysis. Thus, each data becomes point at the map and its number should be about 4~5000.
Prior to clustering, I used the elbow method based on the 'total within cluster sum of squares' to determine appropriate number of cluster. The graph shows approximately 5 graph is good(because the point of inflection is at 5). But the problem is, if I use 5 clusters the number of data would be 18000~20000.
So, is it ok although I set number of clusters about 25? If I do so, I can get the appropriate number of data in target cluster. Of course I know that that method I used to determine the number of clusters is just an auxiliary and analyst's intuition is more important. But since I have a little machine-learning knowledge and still studying, I am not sure if I can determine the number of clusters just by according to number of data I want to extract.
And is it natural that the number of data are uniform to each group in kmeans algorithm? If I use density based clustering method, won't it be that uniform as that of kmeans?
Prior to clustering, I used elbow method based on the total within cluster sum of squaresHow can that be done "prior to clustering"? $\endgroup$