New metaheuristic approaches for optimal stimuli selection

Abstract

In psychological testing, specifically with repeated testing of the same subjects, the optimal choice of used stimuli is essential. In repeated testing, two similar test versions are necessary since the test subject always learns something from the test itself, which means conducting the same test twice would lead to errors in estimates. Other reasons for the need for two test versions include research settings where interest is in the effect of a certain stimuli characteristic. Optimal stimuli selection has received surprisingly little attention from the academic community, and many state-of-the-art optimization methods have not been used to tackle the issue. In this thesis, two new metaheuristic approaches for stimuli selection are proposed. The proposed methods will be based on iterated local search and scatter search metaheuristics. The new methods are compared to the simulated annealing-based method, which has previously seen its fair share of use. The methods were compared with various real-life and simulated datasets in terms of optimality of solutions and used computational time. Resulting test versions were also compared by inspecting descriptive statistics, as the mathematically optimal solution is not guaranteed optimal in terms of practical research use. A comparison of the methods showed that scatter search seemed the best at finding the optimum with the drawback of the computing times getting out of hand in the larger dataset. Simulated annealing showed its strengths as a good all-rounder, while the iterated local search was fasted with the least-optimal solutions.