It may be helpful to run the predictive network management system in a mode such that error between the actual entities and the predictive network management system are measured. This error information can be used during the normal predictive mode in order to help set the above parameters. This begins to remind one of back propagation in a neural network, i.e. the predictive network management system automatically adjusts parameters in response to real output in order to become more accurate.
This calibration mode could be part of normal operation. The error can be tracked simply by keeping track of the difference between the simulated messages and the result of verification queries.