Abstract:

To further increase the training speed of a core vector regression algorithm for large samples, an improved core vector regression algorithm is proposed. This algorithm uses the mapping points of sample data in feature space to obtain the radius of the enclosing ball and makes the radius keep unchanged in the iterative process. By reducing the size of core data set, the training speed is increased. Simulations on prediction of several sets of time series show that on maintaining the same regression accuracy as the core vector regression algorithm, the improved core vector regression algorithm enjoys both less training time and fewer support vectors, so that its efficiency is proved.