About the data
Continuous dependents : v1, v2
Continuous independents : F1, F2, F3, F4
Independent variables F1 and F3 appear to have high importance. And next come F2, F4 variables. Would you please help me to understand more from these results? Please see the attachment.
260 cases with missing data were found.
_________________________________
MARSplines Results:
Dependents: v1, v2
Independents: F1, F2, F3, F4
Number of terms = 9
Number of basis functions = 15
Order of interactions = 2
Penalty = 2.000000
Threshold = 0.000500
GCV error = 0.380457
Prune = Yes
I'd start with
http://statweb.stanford.edu/~tibs/ElemStatLearn/printings/ESLII_print10.pdf
Friedman, who helped develop MARS, is one of the authors of this excellent textbook. (And thanks to those authors who convinced Springer to allow them to have it available freely on the web)
Dear Dr. YILDIZ,
you can use MARS is STATISTICA package program for data miner. Could you please send me the obtained prediction equation belonging to the MARS? I would like to help you in this regard.
Best Regards
Ecevit
Until finding the smallest GCV error, please increase degree!
penalty=2 is the most suitable for MARS!
penalty=-1 can cause over-fitting problems for MARS!
Below you may find R codes for MARS solution with continuous multiresponses
#####################################################
## MULTIVARIATE ADAPTIVE REGRESSION SPLINES ##
## R codes for the prediction of continuous responses more than one ##
######################################################
mydata=read.table("C:/dut.txt", header=T)
str(mydata)
install.packages("earth")
library(earth)
m1=earth(cbind(Y1, Y2)~., data=mydata, penalty=2, degree=1, pmethod="backward", nk=200, nfold=10, keepxy=T)
summary(m1, digits=3, style="max")
evimp(m1)
#####################################################
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis