Success effort and other statistics for performance comparisons in genetic programming

This paper looks at the statistics used to compare variations to the genetic programming method. Previous work in this area has been dominated by the use of mean best-of-run fitness and Koza´s minimum computational effort. This article re-introduces a statistic we name success effort and analyses two methods to produce confidence intervals for the statistic. We then compare success effort and four other performance measures and conclude that success effort is a sometimes more powerful statistic than computational effort and a more desirable measure than the other statistics.