For a Support Vector Machine (SVM), the classification time for an unknown data point is directly proportional to the number of support vectors. Thus, the application of SVM for real-time decision-making, particularly when it uses many support vectors, becomes problematic. This issue may be addressed using expensive hardware if there is no space constraint to install the same. We proposed an algorithm, Self-Organizing Support Vector Machine (SO-SVM) R. Panja, R. K. Mudi and N. R. Pal, SO-SVM: Self-organizing support vector machine, in Smart Trends in Computing and Communications (Springer Nature Singapore, Singapore, 2026), pp. 91–101, that reduces the number of support vectors. Here, we propose an improved version of the algorithm, Re-enforced Self-Organizing Support Vector Machine (RSO-SVM), which is very effective in reducing the number of support vectors without compromising the classification accuracy. The RSO-SVM first clusters the data points using a self-organizing map to find potential class boundary points, which are crucial for determining the separating hyperplane in an SVM. The SVM is then trained using the boundary points selected and the centers of the pure clusters, leading to a reduction in the number of support vectors. The pure cluster centers are included in the training data as they help to make a better representation of the data distribution. The proposed algorithm has been tested on a number of benchmark datasets and is found to reduce the number of support vectors significantly without deteriorating much the classification accuracy.
Panja et al. (Thu,) studied this question.