Anzeige
Erschienen in:
2011 | OriginalPaper | Buchkapitel
A Fast Approximated Evolutionary Approach to Improve SVM Accuracy
verfasst von : Alessandro Perolini
Erschienen in: Research and Development in Intelligent Systems XXVII
Verlag: Springer London
Aktivieren Sie unsere intelligente Suche, um passende Fachinhalte oder Patente zu finden.
Wählen Sie Textabschnitte aus um mit Künstlicher Intelligenz passenden Patente zu finden. powered by
Markieren Sie Textabschnitte, um KI-gestützt weitere passende Inhalte zu finden. powered by
Improving the classification performance is a crucial step of any machine learning method. In order to achieve a better classification Support Vector Machines need to tune parameters and to select relevant variables. To simultaneously perform both targets an embedded approach can be considered. This method consists of a two-layer algorithm where an evolutionary approach handles the solutions and an approximated one evaluates them. The evolutionary search, based on approximated error measures computed on the kernel matrix, allows discovering solutions which have high classification accuracy. The aim of the paper is to verify whether the proposed method is able to find reliable solutions which enhance the classification performance. The proposed method is applied on three real-world datasets using three kernels. In the experiments it is compared against the enclosed Genetic Algorithms and SVMs approach to demonstrate the ability of the approximated method to achieve high classification accuracy in a shorter time.