| Abstrakt: |
The training of the support vector machine (SVM) classifier has high computational complexity and is not suitable for large data classification. Since the classification hyperplane is determined by the support vectors and the other instances do not have an effect on the classifier, a method is introduced that does not use all instances for training. Data set may include inappropriate instances such as noisy and outlier instances. In this paper, a novel method is introduced in which at the first step using the belief function theory, the instances uncertainty such as noisy and outlier instances are identified and discarded, at the second step using the geometric method, called, boundary region, the boundary instances are determined. Finally, at the last step, by using the obtained boundary instances, the training of the SVM classifier is done. In the proposed method BF–BR (Belief Function–Boundary Region), the computational cost of the classification training is reduced without losing classification accuracy. The performance has been evaluated on real world data sets from UCI repository by the tenfold cross validation method. The results of the experiments have been compared with the other methods, which indicate superiority of the proposed method in terms of the number of training instances and training time while good classification accuracy for SVM training is achieved. [ABSTRACT FROM AUTHOR] |
Full text from SpringerLink