I might be missing something, but if you have sets of data where some of it is missing, but you have other data sets where it exists / is complete, then why not train a network with the data you have where it's complete, in order to predict the data where it's not? i.e. a machine learning solution to a machine learning problem.
Hi mam , i am also a Ph.D student doing in agricultural statistics in IASRI. I need to know building models for missing data. Can you send me your mail id. Because I need some help from you mam.