Interpreting canonical coefficients from flexible discriminant analysis when method = mars

[phR]

I’m a beginning R user and a new member of this community. I'm interested in using the mda package to run a flexible discriminant analysis with the mars method. I'm looking for guidance on how to interpret the canonical coefficients I obtain, because the rows are not named like they are when I run an fda using the default method = polyreg. I'm grateful for your help.

coef() when method = mars
                           [,1]                [,2]                  [,3]                   [,4]
[1,] -0.74107385  1.3504962  0.78879242 -1.61802955
[2,]  0.29753770  0.1728769  0.02090793  0.27420208
[3,]  0.21590200 -0.5316448 -0.18151203  0.15314233
[4,]  0.15729606  0.3198006 -0.74194085 -0.47304546

vs.

coef() when method = polyreg

                                                             [,1]                       [,2]                       [,3]                     [,4]
Intercept                   -0.5019719938  -0.070193051     0.55800941  -0.930215219
CP_Psychosis          -0.0683513746  -0.257660684   -0.70681065   0.906489430
CP_Depression       -0.0883615256   0.028353150   -0.28741963  -0.059711890
CP_Anxiety                0.0766194731  -0.151427155   -0.09485747   0.022221503
CP_Oppositional     0.0518207824  -0.075361310  -0.02126432   0.161362909

[Max]

The MARS coefficients are hinge functions and form a segmented regression. Parameters affect subregions of the predictors space so you'd need to know what the cut points are (not just the coefficients). It's very hard to do this unless it is a 1D problem.

For the polynomial functions, it may also be difficult to interpret them (beyond quadratic).

If these are additive models, simple profile plots would be the best way to interpret the patterns. If not, basic model explainers would work well.

[phR]

Ah, I see, thank you for this information. I'll look at MARS in greater detail to determine if I want to pursue this further. Much appreciated!