You're not kooky if this works for you.

Based on your descriptions of the problem, and your statement that this is indeed multivariate rather than univariate, it appears that you want to develop a classifier (i.e. pattern-recognition). You would then test that classifier with input data. One easy way to do so is a statistical construct called a support-vector machine (SVM). They can be implemented almost trivially in the e1071 package for R. An SVM implementation would have the advantage of allowing you to classify individuals rather than groups.

HTH