Evaluating Predictor\'s Accuracy in Evolutionary Algorithms for Dynamic Environments



The addition of prediction mechanisms in Evolutionary Algorithms for dynamic environments is essential to anticipate the changes in the landscape and maximize the adaptability of the algorithm. In previous work, a combination of a linear regression predictor and a Markov chain model was used to enable the EA to anticipate when next change will occur and introduce useful information before it happens. Since the predicted values, in some situations, are not precise, it's necessary to estimate the associated errors. In this paper we introduce a self adaptable parameter calculated using previously observed errors in the linear predictor. This parameter, called  assumes that the linear predictor is not exact and some elasticity must be considered when choosing the moment to introduce useful information into the population. In this work we extend previous studies and nonlinear change periods are introduced to evaluate the prediction accuracy.


Evolutionary Algorithms, Dynamic Environments, Prediction, Markov chains, Linear Regression, Parameter tuning

TechReport Number

TR 2008/004

Cited by

No citations found