A lexicographic cooperative co-evolutionary approach for feature selection

This paper starts with two hypotheses. The first one is that the simultaneous optimization of hyperparameters regulating classifier within a wrapper method, while best subset features being determined, should improve results respect to those obtained pre-parameterized classifier. second solving these problems can be formulated as lexicographic problem, allowing use simple single-objective evolutionary algorithm solve this multi-objective problem. fitness function key importance for such methods. It responsible guiding search towards potentially good solutions and it also consumes most runtime. Having issues in mind, proposes new function, designed minimize runtime avoid over-fitting. Furthermore, execution time quality by procedure depend on some algorithmic hyperparameters: similarity thresholds used when comparing different lexicographically percentage data samples validation during training process. Thus, an experimental analysis has been carried out find adequate values hyperparameters. Finally, cooperative co-evolutionary approach, using proposed paper, tested several datasets belonging University California, Irvine (UCI) repository real high-dimensional datasets, obtaining quite results, compared other state-of-the-art comparison made lexicographically, methodology paper.