Feature and algorithm selection with Hybrid Intelligent Techniques

Abstract

There are many ad hoc feature and algorithm selection techniques, but there also are more methodical approaches. From a theoretical perspective, it can be shown that optimal feature and algorithm selection require an exhaustive search of all possible subsets of features/algorithms. If a large number of features/algorithms are available, this is impractical. In machine learning, the search is often done for a satisfactory set of features/algorithm instead of an optimal set. Feature and algorithm selection have become the focus of some research in hybrid intelligent systems, for which many algorithms and datasets, with tens or hundreds of thousands of variables, are available. Hybrid Intelligent Systems gather researchers who see the need for synergy between various intelligent techniques in solving real problems. In order to minimize the above difficulty, machinelearning algorithms may be used to automatically acquire knowledge for algorithm selection, leading to a reduced need for experts and a potential improvement of performance. That is, the algorithm selection problem can be treated via meta-learning approaches. Several meta-learning approaches have been developed for the problem of algorithm selection. In the first article of this special issue, Kanda et al. select algorithms to solve traveling salesman problem using meta-learning. Meta-learning provides a suitable technique for the selection of the most promising optimization algorithm for a specific instance of the traveling salesman problem. Several experiments illustrate the performance of the meta-learning based approach proposed.