An Improved ARIMA Method Based on Hybrid Dimension Reduction and BP Neural Network

In order to solve the problem that ARIMA model cannot well fit prediction of time series with high dimension and noise, this paper proposes a method based on combination hybrid reduction BP neural network. Taking stock price as an example, proposed takes intraday auxiliary information uses PCA KPCA extract linear nonlinear features it respectively, dimensionally reduced are then used input variable. network was residual error between real value predicted model. Finally, add up closing obtained by for final value. The empirical results show compared model, has better performance fitting accuracy, certain robustness. This can also be extended other practical problems such average temperature port ship flow prediction.