Dear Mr. Shoaib

I am not sure what do you mean by determining the significance of input regarding the observed output.  In case you want to know the significance of specific input(s) in the following model Obs = f(Input1, Input2, ..., Inputn); for non-linear relationship, you can try the GLM (General Linear Model) that available in most statistics software.

I hope this can help you.

Regards,

I Putu EN Kencana

Thanks I Putu for your kind reply. Is it Ok to use linear model for non-linear process?

Significance usually  means with respect to testing a null hypothesis. So this would mean you have a model for the output in response to the input, and you want to calculate the probability of the observed output under this model. Or are you at the stage of exploratory analysis, where you would perhaps start by looking for the correlation between input and output? Making plots and eyeballing should help you thnk what model would be appropriate. Do you expect the process to be stationary, or do you need to figure out change points and so on?

Thanks WM Oxbury for your answer. Let me made my question more clear. I am working on the data driven models for modeling the non-linear rainfall runoff process. As the data driven models simulate the process like Output = f(input1, input2, input 3.....input n). I am trying to find a way to reduce the number of inputs so as to decrease the computational burden and the time. I have already employed the correlation analysis. I wonder if some other method could be applied to reduce/optimize the number of inputs for modeling the non-stationary process.

i guess both linear and non linear process having different procedure. 

juat a suggestion why not to consider an efficiency analysis using the input output analysis rg : CFA