Ensembles of Gradient Boosting Recurrent Neural Network for Time Series Data Prediction

Ensemble deep learning can combine strengths of neural network and ensemble gradually becomes a new emerging research direction. However, the existing methods either lack theoretical support or demand large integrated models. To solve these problems, in this paper, Ensembles Gradient Boosting Recurrent Neural Network (EGB-RNN) is proposed, which combines gradient boosting framework with three types recurrent models, namely Minimal Gated Unit (MGU), (GRU) Long Short-Term Memory (LSTM). RNN model used as base learner to integrate an learner, through way boosting. Meanwhile, for ensuring fit data better, Step Iteration Algorithm designed find appropriate rate before models being integrated. Contrast trials are carried out on four time series sets. Experimental results demonstrate that number integration increasing, performance EGB-RNN tend converge best degree vary It also shown statistical perform better than six baselines.