I have a large set of records with each record containing $N$ variables $v_i$ (say $N=20$).
I can transform these variables $v_i$ to be in the range [0,1] and am able to use a simple loss function $L = \sum_i v_i^2$. After that I know that there are two large clusters of points with small $L$ and large $L$. There are also few points in between these clusters. I can successfully assign records to these clusters.
R functionality would I use to get a parametric representation (in the variables $v_i$) of the surface of the "large $L$ cluster", preferably as a function of $L$?
(I know I could train a neural network on my data to give me a similar answer, but I would like to stay away from that for this application.)