In order to predict stock index, an empirical mode decomposition (EMD) based on support vector machine (SVM) ensemble learning paradigm was proposed. Firstly, the original stock index series were first decomposed into a finite number of independent intrinsic mode functions (IMFs), with different frequencies. Then the IMFs were composed into high-frequency sequence, low-frequency sequence, trend series. Secondly, based on the analysis of Lemple-Ziv complexity, the right kernel functions were chosen to build different SVM respectively to predict each IMF. Then, the sum of each forecasting value of equal weighted will be the final prediction. Finally, we select Shanghai and Shenzhen 300 index (CSI 300 Index) as the empirical research and the results demonstrate effectiveness and attractiveness of the proposed EMD-based SVM model compared with SVM based on fish-swarm algorithm (FSA) optimization.
展开▼