An incremental dual nu-support vector regression algorithm

Abstract

Support vector regression (SVR) has been a hot research topic for several years as it is an effective regression learning algorithm. Early studies on SVR mostly focus on solving large-scale problems. Nowadays, an increasing number of researchers are focusing on incremental SVR algorithms. However, these incremental SVR algorithms cannot handle uncertain data, which are very common in real life because the data in the training example must be precise. Therefore, to handle the incremental regression problem with uncertain data, an incremental dual nu-support vector regression algorithm (dual-v-SVR) is proposed. In the algorithm, a dual-v-SVR formulation is designed to handle the uncertain data at first, then we design two special adjustments to enable the dual-v-SVR model to learn incrementally: incremental adjustment and decremental adjustment. Finally, the experiment results demonstrate that the incremental dual-v-SVR algorithm is an efficient incremental algorithm which is not only capable of solving the incremental regression problem with uncertain data, it is also faster than batch or other incremental SVR algorithms.