Simulation performance prediction methods make possible the realization of performance improvement potentials of Parallel Discrete Event Simulation (PDES) methods, important in the analysis of complex systems and large-scale networks. Currently, high performance execution environments (emerging clusters and computing clouds) advance the development of quality/cost analysis capabilities of performance prediction methods. In this paper, for the evaluation and management of prediction correctness/cost, the efficacy, efficiency and effectiveness coefficients and improvement operations are defined for predictions. The performance coefficients and improvement operations are embedded in the rough-set-modeling and learning process and presented as an enhancement approach of the conventional Coupling Factor Method (CFM). A case study based on the CFM analysis of PDES of a closed queuing network model is presented. In the example, after rough-modeling and train-and-test analysis, the correctness/cost evaluation and effectiveness improvement operations are shown for series of predictions and the feedback connection to modeling refinement phase is demonstrated too.