Time Series Forecasting Model with Error Correction by Structure Adaptive RBF Neural Network

A hybrid methodology is proposed to take advantage of the unique strength of autoregressive integrated moving average (ARIMA) and RBF (radial basis function) neural networks in linear and nonlinear modeling, which is an error correction method to create synergies in the overall forecasting process. ARIMA model is used to generate a linear forecast in the first stage, and then RBFN is developed as the nonlinear pattern recognition to correct the estimation error in ARIMA forecast. A dynamic clustering algorithm is developed to optimize the network structure, which makes the RBFN adapt to the specified training set, reduces computation complexity and avoids overfitting. With two real datasets, in terms of forecasting accuracy, empirical results evidently show that the hybrid model outperforms noticeably ARIMA and RBFN model used in isolation

[1]  Anders Krogh,et al.  Neural Network Ensembles, Cross Validation, and Active Learning , 1994, NIPS.

[2]  Stephen A. Billings,et al.  On-line Supervised Adaptive Training Using Radial Basis Function Networks , 1996, Neural Networks.

[3]  Ibrahim El-Amin,et al.  Artificial neural networks as applied to long-term demand forecasting , 1999, Artif. Intell. Eng..

[4]  Martin Burger,et al.  Error Bounds for Approximation with Neural Networks , 2001, J. Approx. Theory.

[5]  James J. Carroll,et al.  Approximation of nonlinear systems with radial basis function neural networks , 2001, IEEE Trans. Neural Networks.

[6]  H. Brian Hwarng,et al.  Insights into neural-network forecasting of time series corresponding to ARMA(p,q) structures , 2001 .

[7]  Timo Teräsvirta,et al.  The combination of forecasts using changing weights , 1994 .

[8]  Michael Y. Hu,et al.  A simulation study of artificial neural networks for nonlinear time-series forecasting , 2001, Comput. Oper. Res..

[9]  Vassilis S. Kodogiannis,et al.  Forecasting Financial Time Series using Neural Network and Fuzzy System-based Techniques , 2002, Neural Computing & Applications.

[10]  Katsuyuki Hagiwara,et al.  Regularization learning, early stopping and biased estimator , 2002, Neurocomputing.

[11]  James V. Hansen,et al.  Time-series analysis with neural networks and ARIMA-neural network hybrids , 2003, J. Exp. Theor. Artif. Intell..

[12]  Y. Abe,et al.  Fast computation of RBF coefficients for regularly sampled inputs , 2003 .

[13]  J. V. Hansen,et al.  Forecasting and recombining time-series components by using neural networks , 2003, J. Oper. Res. Soc..

[14]  Noureddine Zerhouni,et al.  Recurrent radial basis function network for time-series prediction , 2003 .

[15]  Guoqiang Peter Zhang,et al.  Time series forecasting using a hybrid ARIMA and neural network model , 2003, Neurocomputing.

[16]  Chiung-Hsin Tsai,et al.  Deadzone compensation based on constrained RBF neural network , 2004, J. Frankl. Inst..

[17]  An-Sing Chen,et al.  Regression neural network for error correction in foreign exchange forecasting and trading , 2004, Comput. Oper. Res..

[18]  Manoochehr Ghiassi,et al.  A dynamic architecture for artificial neural networks , 2005, Neurocomputing.

[19]  Guoqiang Peter Zhang,et al.  Neural network forecasting for seasonal and trend time series , 2005, Eur. J. Oper. Res..