Ensemble Learning Based On Neural Networks And Fuzzy Cognitive Map And Its Applications

In recent years, forecasting nonlinear time series is a hot research topic in the field of scientific research. To overcome the drawbacks of original ensemble methods, this paper improve the ensemble learning from four different aspects:ensemble algorithm; parameter adjustment; ensemble modelling and the selection of base learner for ensemble modelling.In the first part, we choose BP neural networks as the based learners for ensemble algorithm:AdaBoost.RT. Meanwhile, from the perspective of the selection of training data, we propose two novel strategies:multiple modeling strategy and global modeling strategy.Secondly, it is quite important for AdaBoost.RT algorithm to set the exact value of threshold. In this chapter, we propose a novel paradigm to self-adaptive learning for threshold in this ensemble algorithm.Thirdly, we propose a novel selective ensemble method based on Elman neural network and artificial immune algorithm. In this way, we forecast nonlinear time series based on selective ensemble methods.Last but not the least, we combine fuzzy cognitive maps with fuzzy-c-means clustering algorithm to build a novel hybrid model and then regard this hybrid model as the base learners for ensemble models to predict the nonlinear time series.