Fighting Program Bloat with the Fractal Complexity Measure

摘要

The problem of evolving decision programs to be used for medical diagnosis prediction brought us to the problem, well know to the genetic proramming (G) community - the tendency of programs to grow in length too fast. While searching for a solution we found out that an appropriately defined fractal complexity measure can differentiate between random and nonrandom computer programs by measuring the fractal structure of the computer programs. Knowing this fact, we introduced the fractal measure #alpha# in the evaluation and selection phase of the evolutionary process of decision program induction, which resulted in a significant program bloat reduction.