Assuming you're not interested in the more high-end approaches like random forests or model averaging, you might want to try a mixed effects model where you include database as a random effect.

Thank you for the suggestion. Indeed we have been working with model averaged methods, in the field of credit risk management, and they have proved valid. I am now looking for competing methods to benchmark with. do you know applications of the mixed modelling approach you suggest in predictive problems?