A multi-objective approach for the ( α, β )- k -feature set problem using memetic algorithms
Francia Jiménez, Claudio Sanhueza, Regina Esther Berretta, Pablo Moscato · Proceedings of the Genetic and Evolutionary Computation Conference Companion · 2017
Nowadays, the ceaseless data gathering in science and technology is bringing new challenges. Companies use the collected data to create new digital services and products. These services rely on innovations in data mining algorithms. The combinatorial search required to select a subset of features that best describe classes in a dataset is a challenging problem in this research field. The (α, β) - k- Feature Set Problem is a mathematical model proposed for addressing this task. In its most common optimization variant, the problem has always been to find the minimum number of features for given fixed values of α and β that satisfy the requirements of the model. However, the relation between the α and β parameters and the number of features is unknown. In the literature, multi-objective approaches have been used, with great success, to address problems that require optimizing several objectives simultaneously. In this study, we propose a novel multi-objective approach for solving the (α, β) - k- Feature Set Problem using memetic algorithms. We study and evaluate different local searches and initialization procedures using six well-known datasets. Our results show that the clustering-based local search heuristic has a positive impact on the quality of the solutions.