# Parameter Optimization of Nu-SVR for Electric Market Price

Hi

i am preparing forecasting model by nu-SVR model for electric market price as academic study. I used evolutionary parameter optimization (parallel) to optimize nu-SVR model. Inside parameter optimization operator i used split validation (regression) as inner operator. Split validation parameters are relative split, split ratio 0.9, sampling type shuffled sampling. main criterion of performance operator is root mean squared error. Optimization parameter resuts are like that:

SVM.C = 374.0283602994765

SVM.nu = 0.43167817384702967

SVM.epsilon = 1.0E-4

SVM.gamma = 9.684928217146895E-4

Now i want to learn this questions to get better result:

-What can be main criterion of performance operator?

-How length can the parameters of optimizations be? I used for C 0-600, for nu 0-0.5, for epsilon 0.0001-0.01, for gamma 0-0.001.

-What can be parameters of parameter optimization? I used switching mutation, tournament selection (tournament fraction=0.25), crosover prob=0,6, population size=6 max generations=70

My results for these parameters mentioned above are

PerformanceVector [

*****root_mean_squared_error: 14.059 +/- 0.000

-----relative_error: 8.86% +/- 11.22%

-----root_relative_squared_error: 0.386

-----squared_error: 197.655 +/- 386.029

-----squared_correlation: 0.852

I really wonder about optimization of support vector regression and its evaluation. According to this result, my model is sufficient? How can i improve the model? I know this question can not resulted easily? but i expect a guide to better the model.

i am preparing forecasting model by nu-SVR model for electric market price as academic study. I used evolutionary parameter optimization (parallel) to optimize nu-SVR model. Inside parameter optimization operator i used split validation (regression) as inner operator. Split validation parameters are relative split, split ratio 0.9, sampling type shuffled sampling. main criterion of performance operator is root mean squared error. Optimization parameter resuts are like that:

SVM.C = 374.0283602994765

SVM.nu = 0.43167817384702967

SVM.epsilon = 1.0E-4

SVM.gamma = 9.684928217146895E-4

Now i want to learn this questions to get better result:

-What can be main criterion of performance operator?

-How length can the parameters of optimizations be? I used for C 0-600, for nu 0-0.5, for epsilon 0.0001-0.01, for gamma 0-0.001.

-What can be parameters of parameter optimization? I used switching mutation, tournament selection (tournament fraction=0.25), crosover prob=0,6, population size=6 max generations=70

My results for these parameters mentioned above are

PerformanceVector [

*****root_mean_squared_error: 14.059 +/- 0.000

-----relative_error: 8.86% +/- 11.22%

-----root_relative_squared_error: 0.386

-----squared_error: 197.655 +/- 386.029

-----squared_correlation: 0.852

I really wonder about optimization of support vector regression and its evaluation. According to this result, my model is sufficient? How can i improve the model? I know this question can not resulted easily? but i expect a guide to better the model.

0

## Answers

1,869UnicornBest, Marius

9Contributor III will rebuild the model according to your instructions.