Abstract
Currently advancement in computational power leads to study of physical processes through deterministic computer simulated experiments (CSE). Normally CSEs are time consuming and computationally expensive to run. Hence a selection of proper design to run CSE is very important. The space filling designs, like Latin hypercube design (LHD) which aim to spread the design points over the region of interest are commonly used to generate the output of CSE. The optimal design for a given problem is obtained using search algorithms with a pre-specified optimality criterion. Searching for the optimal design is computational burden especially when the dimension of the problem is large. This paper aims to investigate the relation between the most commonly used optimality criterion namely ɸ p criterion and prediction accuracy of the best designs obtained from search algorithms. The results show that prediction capability would not be affected by different values of criteria as long as we terminate search algorithms within 20% of the threshold distance from the global minimum of ɸ p criterion for small dimensions of problem while a shorter threshold distance, as low as 10% from the global minimum is required for larger dimensions of problem.

This publication has 11 references indexed in Scilit: