METHODS FOR REDUCING SEARCH AND EVALUATING FITNESS FUNCTIONS IN GENETIC ALGORITHMS FOR THE SNAKE-IN-THE-BOX PROBLEM by JOSHUA

This thesis explores the use of Genetic Algorithms in approaching the Snake-In-The-Box problem in dimension 8. It discusses methods for improving the solutions found by reducing the representation space for the problem. It presents a representation scheme called FrequencyBased Transition Reassignment (FBTR), which creates a unified interpretation of individuals to prune the search space. FBTR is compared to a standard transition-based representation to determine its effectiveness. It is also compared to a canonical representation that was presented by K. J. Kochut in 1996 which is a different technique meant to reduce the search space. In addition, this thesis introduces the concept of a snake blocker and identifies methods for dealing with snake blockers. These methods are evaluated for their impact on the GA as a whole. Furthermore, fitness functions are explored in great detail and a variety of components to supplement the length of the longest snake in the chromosome, are suggested and evaluated. These components include tightness, skin density, selective skin density, and target distribution. INDEX WORDS: Snake-In-The-Box, Genetic Algorithms, Hypercube, Snake Problem, Frequency-Based Transition Reassignment, FBTR, Tightness, Skin Density, Selective Skin Density, SSD, Target Distribution, Snake Blockers, Restricted Random Initialization, RRI, Restricted Heuristic Initialization, RHI, Canonical Representation METHODS FOR REDUCING SEARCH AND EVALUATING FITNESS FUNCTIONS IN GENETIC ALGORITHMS FOR THE SNAKE-IN-THE-BOX PROBLEM