Variable selection for multivariate time series prediction with neural networks

This paper proposes a variable selection algorithm based on neural networks for multivariate time series prediction. Sensitivity analysis of the neural network error function with respect to the input is developed to quantify the saliency of each input variables. Then the input nodes with low sensitivity are pruned along with their connections, which represents to delete the corresponding redundant variables. The proposed algorithm is tested oil both computer-generated time series and practical observations. Experiment results show that the algorithm proposed outperformed other variable selection method by achieving a more significant reduction in the training data size and higher prediction accuracy.