System Selection Based on Controlled Risk

The studies of using computer simulation to address stochastic systems selection problem prevail currently. Ranking and selection procedures are statistical procedures that can be used in the field of computer simulation to select the best system or the subset that includes the best system based on a performance metric of the interest with a guaranteed correct selection. The best system is defined in the majority of selection problems as a system with the largest (or smallest) mean value. However, mean is only a measure of the average behavior. Risk, which influences the stability of a system to a large extend, may not be controlled by controlling the mean. In this study, we focus on developing stochastic system selection procedures that consider the mean and the risk of systems simultaneously. Variance is used as a measure of risk. Three different ranking and selection procedures are designed to satisfy the decision maker's preference via i) mean-variance measure, ii) mean with variance constraint measure, and iii) risk-adjusted mean measure. The validation of each procedure is proved, and the experiments are illustrated at the end of this thesis.