The use of metaheuristics for feature selection and the derivation of diagnostic rules

Abstract

[ACCESS RESTRICTED TO THE UNIVERSITY OF MISSOURI AT AUTHOR'S REQUEST.] This dissertation includes four chapters that discuss 1) the history of metaheuristics, 2) the development of a genetic algorithm for feature selection, 3) the development of a genetic algorithm for deriving psychiatric diagnoses and 4) a demonstration of deriving shortened diagnostic rules for alcohol use disorder that optimally agree with the DSM-5. The first chapter offers an overview of novel developments in the metaheuristics literature, along with suggestions for future developments. The second and third chapters of this dissertation 1) propose new algorithms that can handle search spaces that are not accessible by current algorithms and 2) examine each component of the proposed algorithms to identify subordinate heuristics that are essential for the success of the algorithm. The final chapter utilizes information obtained from the previous two chapters to assess the performance of an algorithm for deriving diagnostic rules in a supervised learning context.