Genetic Algorithms in Search, Optimization and Machine Learning
Genetic Algorithms in Search, Optimization and Machine Learning
| Hi-index | 0.00 |
The problem treated is the comparison of the performance of two different fitting index, defined based in Information Theory criteria (Kullback-Leibler and Akaike) for the genetic algorithm. Samples of known models are generated with the IPF modified algorithm. As the true model is known, the performance is assessed through the percent of true edges of the true model in the best model selected by the algorithm. An experiment is designed to select the best combination of parameters of the considered genetic algorithm, and nine known models of different complexities are used to assess the performance of the genetic algorithm with each of the two fitting criteria. As results the genetic algorithm with the convex fitting index (CFI) find more or equal (equal only in one case) number of correct edges than the mixed fitting index (MFI), the better performance of the genetic algorithm occurs always selecting 60 percent of the best population individuals for reproducing the next population, the edges of the sparse models are more difficult to find for any of the fitting indexes than the edges of the medium and dense models. The more complex the models are, the better the performance of the algorithm is. The significant factors were: the fitting indexes, the percent of the best individuals selected to reproduce, and the complexity of the models.