A binary-encoded tabu-list genetic algorithm for fast support vector regression hyper-parameters tuning

The selection of hyper-parameters in support vector machines for regression (SVMr) is an essential step in the training process of these learning machines. Unfortunately, there is not an exact method to obtain the optimal values of SVM hyper-parameters. Therefore, it is necessary to use a search algorithm in order to find the best set of hyper-parameters. Grid Search is the most commonly used option to perform such a hyper-parameters search, though other possibilities based on evolutionary computation algorithms have been proposed in the literature. In this paper we analyze the use of a standard genetic algorithm with binary encoding, which allows a fast exploration of the hyper-parameters space. We include a kind of tabu-list in the proposed algorithm, where we keep the last individuals generated by the genetic algorithm to avoid re-training of the SVMr with them. This technique allows a good improvement of the SVMr training time respect to the grid search approach, while keeping the machine accuracy almost unaltered.