06-05-2018 11:10 AM
I am looking to model data that appears to follow the exponential distribution with a mean of 1.8, as indicated by the screenshots from the CAPABILITY procedure.
Knowing this, how would one go on about trying to build a predictive model for this data? I narrowed down the number of variables to use with the Rapid Predictive Modeler (I do not have Enterprise Miner installed on my computer) and I have tried using the MODEL procedure to do a non-linear modeling of the data but the fit is very poor with an r-square of 0.0674 and a root MSE of 1.8 (to give perspective, the mean of the data is 1.8). Any advice on this topic would be greatly appreciated!
PROC MODEL DATA=stdize_data; PARAMETERS w_1 0 w_2 0 w_3 0 w_4 0 lambda 1; BOUNDS w_1 >= 0, w_2 >= 0, w_3 >= 0, w_4 >= 0, lambda > 0; RESTRICT w_1 + w_2 + w_3 + w_4 = 1; CER = lambda * exp(-lambda * (w_1*x_1 + w_2*x_2 + w_3*x_3 + w_4*x_4)); FIT CER / MAXITER=10000 OUT=model_results OUTALL; RUN;
06-05-2018 03:10 PM
I guess I don't understand what you mean by "how do I build a predictive model." If there are no covariates, then the expected value of your model is 1.8 and the model
Y = 1.8
is the linear model that fits the data.
In your PROC MODEL statement, you have X1-X4, which appear to be explanatory variables. But if these are covariates, then you should not have used PROC CAPABILITY to fit the response, you should have used a regression procedure such as PROC GENMOD.
Please explain whether this is a univariate model or a regression problem. If regression, please describe the data.