An efficient online learning approach for support vector regression

In this paper, an efficient online learning approach is proposed for Support Vector Regression (SVR) by combining Feature Vector Selection (FVS) and incremental learning. FVS is used to reduce the size of the training data set and serves as model update criterion. Incremental learning can “adiabatically” add a new Feature Vector (FV) in the model, while retaining the Kuhn-Tucker conditions. The proposed approach can be applied for both online training & learning and offline training & online learning. The results on a real case study concerning data for anomaly prediction in a component of a power generation system show the satisfactory performance and efficiency of this learning paradigm.