You forgot to also mention temperature, road surface properties, frame dimensions and angles, lots of wheel parameters, fork parameters, lots of material properties, etc. I'm a ME and do a lot of multivariate, nonlinear analysis and modeling. As a practical matter, I don't think the OP's thesis holds water. A workable approach would be to build (solids, FEA) computer models of bikes+components with some physical testing to calibrate the models, then put people on bikes and poll them, and then correlate the poll results to the properties of the bikes. One would expect clusters based on how people use bikes, e.g., touring vs. racing, but who knows.