A Three-stage SVM Ensemble Algorithm for Chaotic Time Series Prediction

Inspired by the so-called "divide-and-conquer" principle that is often used to attack a complex problem by dividing it into simpler problems, a three-stage SVM ensemble algorithm is proposed to improve its prediction accuracy and generalization performance for chaotic time series. In the first stage, Fuzzy C-means clustering algorithm is adopted to partition the input dataset into several subsets. Then, in the second stage, SVMs with composite kernels that best fit partitioned subsets are constructed respectively, which hyperparameters are adaptively evolved by the particle swarm optimization (PSO) algorithm, and in the third stage, a fuzzy synthesis algorithm is employed to combine the outputs of submodels to obtain the final output, in which the degrees of memberships are generated by the relationship between a new input sample data and each subset center. Simulation results on a chaotic benchmark time series indicate that the presented algorithm shows good prediction performance compared to the other existing algorithms for the time series prediction task considered in this paper.