Machine Learning Idea – A sort of generalization is putting a line between points in a curve. Linear Part Of a Maclaurin Series f(x) ~= f(0)+f'(0)*x+… This Suggest you ?can transform the multidimensional input to a single parameter.Compare the linear approximation with y = model(x) between points. Should give a more stable model().