ABSTRACT
Success Rate (SR) is a statistic straightforward to use and interpret, however a number of non-trivial statistical issues arises when it is examinated in detail. We address some of those issues, providing evidence that suggests that SR follows a binomial density function, therefore its statistical properties are independent of the flavour of the Evolutionary Algorithm (EA) and its domain. It is fully described by the SR and the number of runs. Moreover, the binomial distribution is a well known statistical distribution with a large corpus of tools available that can be used in the context of EC research. One of those tools, confidence intervals (CIs), is studied.
- L. D. Brown, T. T. Cai, and A. Dasgupta. Interval estimation for a binomial. Statistical Science, 16:101--133, 2001.Google ScholarCross Ref
- S. Christensen and F. Oppacher. An analysis of koza's computational effort statistic for genetic programming. In EuroGP'02, pages 182--191, London, UK, 2002. Springer-Verlag. Google ScholarDigital Library
- M. Walker, H. Edwards, and C. H. Messom. Confidence intervals for computational effort comparisons. In EuroGP, pages 23--32, 2007.. Google ScholarDigital Library
Index Terms
Confidence intervals of success rates in evolutionary computation
Recommendations
Confidence intervals for steady state availability of a system with exponential operating time and lognormal repair time
Long-run availability of a system assuming lognormal repair times and exponential failure times is considered. With many repairable systems, the time to repair is well characterized by a lognormal distribution. However, due to the difficulties ...
Approximate confidence intervals for one proportion and difference of two proportions
Constructing a confidence interval for a binomial proportion or the difference of two proportions is a routine exercise in daily data analysis. The best-known method is the Wald interval based on the asymptotic normal approximation to the distribution ...
Asymptotic confidence intervals for Poisson regression
Let (X,Y) be a R^dxN"0-valued random vector where the conditional distribution of Y given X=x is a Poisson distribution with mean m(x). We estimate m by a local polynomial kernel estimate defined by maximizing a localized log-likelihood function. We use ...
Comments