Hi dear
How we can improve the result of lssvm in regression model, I find the best result for 81 sample with 7 variables as well R2=0.84. Is there any way to improve this result in lssvm toolbox in matlab. I use both of the parameters search method in this toolbox, (Simplex,Gridsearch) but there was no different between result.
The Best
Hi,
If you absolutely need to LSSVM, you can use an Automated Algorithm Configuration method, like SMAC. It is very efficient for finding the best parameter settingfor a given algorithm.
Then the question is if you using LSSVM is not mandatory, you should look for new algorithms, like kernel based algorithm like an svm with a rbf kernel.
https://github.com/belkhir-nacim/ac_pysmac/
http://www.cs.ubc.ca/labs/beta/Projects/SMAC/
Hi dear Nacim
About your first question, there is no forces to use libsvm or lssvm, I try to use nonlinear methods for modeling 83 sample with 7 variables that their correlation coefficient in MLR model was 0.76, I try to apply svr by libsvm but I cant find valuable and better result than MLR model. In next step I use lssvm and find better result with R2:0.84.
In lssvm toolbox, I use simplex (also grid search) and RBF kernal to construct regression model. Their is no important which svm algorithm or toolbox was used, I wanna to find model with the highest correlation and lowest error.
I try to check your solution to find better result.
The best
Mohamad
An alternative would be to use a random forest or gradient boosting instead of svm.
- Badges
- Science topic