Pre-processing inputs for optimally-configured time-delay neural networks

A procedure for pre-processing non-stationary time series is proposed for modelling with a time-delay neural network (TDNN). The procedure stabilises the mean of the series and uses a fast Fourier transform to determine the TDNN input size. Results of applying this procedure on five well-known data sets are compared with existing hybrid neural network techniques, demonstrating improved prediction performance.

[1]  S. Weisberg.  Plots, transformations, and regression , 1985 .

[2]  Geoffrey E. Hinton,et al.  Phoneme recognition using time-delay neural networks , 1989, IEEE Trans. Acoust. Speech Signal Process..

[3]  Bernd Freisleben,et al.  Nonstationarity and data preprocessing for neural network predictions of an economic time series , 2000, Proceedings of the IEEE-INNS-ENNS International Joint Conference on Neural Networks. IJCNN 2000. Neural Computing: New Challenges and Perspectives for the New Millennium.

[4]  Fang-Mei Tseng,et al.  Combining neural network model with seasonal time series ARIMA model , 2002 .

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

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

[7]  D. Cox,et al.  An Analysis of Transformations , 1964 .

[8]  Marie Cottrell,et al.  Neural modeling for time series: A statistical stepwise method for weight elimination , 1995, IEEE Trans. Neural Networks.