Enhancing the Performance of SVM on Skewed Data Sets by Exciting Support Vectors
In pattern recognition and data mining a data set is named skewed or imbalanced if it contains a large number of objects of certain type and a very small number of objects of the opposite type. The imbalance in data sets represents a challenging problem for most classification methods, this is because the generalization power achieved for classic classifiers is not good for skewed data sets. Many real data sets are imbalanced, so the development of new methods to face this problem is necessary. The SVM classifier has an exceptional performance for data sets that are not skewed, however for imbalanced sets the optimal separating hyper plane is not enough to achieve acceptable results. In this paper a novel method that improves the performance of SVM for skewed data sets is presented. The proposed method works by exciting the support vectors and displacing the separating hyper plane towards majority class. According to the results obtained in experiments with different skewed data sets, the method enhances not only the accuracy but also the sensitivity of SVM classifier on this kind of data sets.
KeywordsSVM skewed data sets imbalanced data sets SMOTE
Unable to display preview. Download preview PDF.
- 2.Arbach, L., Reinhardt, J., Bennett, D., Fallouh, G.: Mammographic Masses Classification – Comparison Between Backpropagation Neural Network (BNN), K Nearest Neighbors (KNN), and Human Readers. In: IEEE Canadian Conference on Electrical and Computer Engineering (CCECE 2003), vol. 3, pp. 1441–1444. IEEE, Washington, DC (2003)Google Scholar
- 9.Koknar-Tezel, S., Latecki, L.: Improving SVM Classification on Imbalanced Data Sets in Distance Spaces. In: 9th IEEE International Conference on Data Mining (ICDM 2009), pp. 259–267. IEEE, Washington, DC (2009)Google Scholar
- 16.Vapnik, V.N.: Statistical Learning Theory. Wiley-Interscience (1998)Google Scholar
- 17.Veropoulos, K., Campbell, C., Cristianini, N.: Controlling the Sensitivity of Support Vector Machines. In: 16th International Joint Conference on Artificial Intelligence (IJCAI 1999), pp. 55–60 (1999)Google Scholar