Using Ordinal Optimization Approach to Improve Efficiency of Selection Procedures

Indifference-zone selection procedures have been widely studied and applied to determine the sample sizes for selecting a good design among k alternative designs. However, efficiency is still a key concern for using simulation to solve ranking and selection problems. Ordinal optimization has emerged as an effective technique to improve efficiency of simulation and optimization. In this paper, we incorporate the concept of ordinal optimization with ranking-and-selection methodology and propose using a normal approximation to estimate the probability of correct selection. The proposed procedure takes into account not only the sample variances but also the difference of sample means when determining the sample sizes. Furthermore, the procedure is valid with the variance reduction technique of common random numbers. An experimental performance evaluation demonstrates the efficiency of the new procedure.