A new incremental learning method based support vector regression for system modeling

This paper proposes a new incremental learning method based on weighted constraints support vector regression, called IL-WCSVR. First, a modified nearest neighbor clustering method is used to partition the whole input space into several sub-clusters. Then, according to the distance distinguish condition, all the samples in sub-cluster are identified to form the candidate support vector set. For each new sample added, it was clustered to an individual candidate support vector subset where the cluster center is the closest to the new sample. Finally, in the WCSVR learning process, the object function is modified by introducing a weight for each new sample and the corresponding regularization term. The experiment shows prominence of our proposed incremental method in comparison with the standard SVR.