A0472
Title: Genetic versus controlled approximate algorithms for regression model selection
Authors: Georgiana-Elena Pascaru - Alexandru Ioan Cuza University of Iasi (Romania)
Cristian Gatu - University of Iasi (Romania) [presenting]
Erricos Kontoghiorghes - Cyprus University of Technology and Birkbeck University of London, UK (Cyprus)
Abstract: Algorithms for regression model selection are compared in terms of execution times and quality of obtained solution. Specifically, the recently introduced heuristic algorithm (HBBA) and implemented in the R package ``lmSubsets'' is compared to the genetic algorithms (GA). The targeted GAs are ``gaselect'', ``kofnGA'', and ``glmulti'' that are also implemented as R packages. The HBAA yields solutions having relative errors with respect to the optimum that lie within a given tolerance. Thus the quality of HBAA solutions can be properly assessed. The GA obtain also solutions that are not optimal and do not provide any information as how far they are from the optimum. The aim is to compare the HBBA and GA. Specifically, the comparison (a) investigates the maximum problem size that can be tackled by the HBBA and GA within a given reasonable computing time; (b) determines the average relative errors of the GA solutions, say $\tau_{GA}$, when compared to the optimum solutions which are obtained by HBBA with zero tolerance; and (c) assess the execution times of the GA and the HBBA with tolerance controlled by $\tau_{GA}$.
