WSEAS Transactions on Advances in Engineering Education
Print ISSN: 1790-1979, E-ISSN: 2224-3410
Volume 10, 2013
A Comparative Study of SVM Models for Learning Handwritten Arabic Characters
Authors: , ,
Abstract: In order to select the best SVM model for a specific machine learning task, a comparative study of SVM models is presented in this paper. We investigate the case of learning handwritten Arabic characters and we make use of tabu search metaheuristic in order to scan a large space of SVM models including multi-class scheme (one-against-one or one-against-all), SVM kernel function and kernel parameters. These parameters have a great influence on final performance of the classifier and also on computation time. This work has involved the creation of a complete offline system for learning handwritten Arabic characters, generating a corpus of 4840 Arabic characters in their different positions (beginning, middle, end and isolated). Based on some theoretical interpretations and simulation results, the effect of SVM model on prediction rate and CPU time is discussed.
Search Articles
Keywords: Character recognition, handwritten Arabic character recognition, Support Vector Machines, model selection, tabu search