UQWorld

Reliability Analysis with Support Vector Regression(SVR)

Hello Everyone, I really need some clarification relating to carrying out reliability analysis in UQLab having develop a metamodel with Support Vector Regression(SVR). From what I noticed, you typically do not get an expression showing relationship between dependent and independent variables like linear regression for SVR. How can reliability analysis be carried out using the reliability module such as SORM, MCS in this regard? Appreciate your help on this.

Dear Aghatise Okoro,

I am not sure I understood your first question but the yes SVR expansion relates the observations in the feature space rather than the input variables themselves. The easiest way to use support vector regression for reliability analysis is to use the built approximation in lieu of the original model. Once an approximation is built you can then use any reliability method, be it Monte Carlo, SORM or subset simulation, in combination to the SVR model.
Is there any specific reason you are using support vector regression? If not, I would recommend having a look at Kriging and more specifically AK-MCS which is currently implemented in UQLab and allows you to adaptively build your metamodel for the purpose of reliability analysis.

Cheers,
Moustapha

Thank you so much Moustapha - appreciate your response.