Hello,
I would like to perform feature selection in a set of features that are
used for regression. Especially, those features correspond to the previous
day values (e.g Lag24,Lag25,Lag26...) where lag24 is the value 24 hour
before. The target variable y is the value at the current time (Using past
day features in order to predict the next day). I am currently using SVM
from the e1071 package. However, I found that when I remove some features
the svm performance is increased? Is there any way so to do feature
selection using the SVM? (1). Also I have tried to use the glmnet package
for doing regression but with no luck. The purpose for using the glmnet was
the LASSO penalizing on the model. Can I do something similar using e1071?
(2) . I am not using any penalizing in e1071 so maybe this is an issue.
Also could you please list me 2-3 packages used for non-linear regression.
(3)
Currently I am aware of:
e1071
penalizedSVM
randomForrest
RSNNS (elman, jordan neural networks)
forecast (For using ARIMA)
glmnet (No luck)
I have tried may of these but without very good results even if my data
have a periodicity (25% Mean Relative Absolute Error).
For feature selection until now I use the corrgrams function that returns
the correlation of the features.
My Questions have the symbol ( Question Id).
Thank you for your support.
[[alternative HTML version deleted]]
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
