23 Taxonomy of a Diverse Collection of String Optimization Problems

This study builds on previous work that classifies evolutionary computation problems via graph-based taxonomic techniques. A new collection of graphs that permit the measurement of small population effects are employed as well as a new distance measure that incorporates information on the variability of problem hardness into the taxonomic process. The new set of graphs and improved distance measure on problems are tested on a diverse set of string-based optimization problems including two new types of problems: fragmentation and K-max. The familiar SAW, one-max, and royal road problems are also included. It is found that problem difficulty, although explicitly removed from the information used to classify the problems, is nevertheless detected by the taxonomic technique. The problems appearing in earlier studies are shown to retain their diversity of behavior, confirming that the improvements to the problem classification system did not degrade a valuable feature of the earlier system.