Extracting rules from neural networks for time series prediction
Nguyen Hoai Le, Duong Tuan Anh · 2014
A significant limitation of neural networks is that the representations they learn are usually incomprehensible to humans. There have been a number of research works that focused on how to extract rules from trained neural networks. Recently, Kamruzzaman et al. have developed an efficient algorithm, called REANN, for extracting rules from trained neural networks for classification problem. Given a trained network, REANN produces a set of rules that approximates the function represented by the network. In this paper, we investigate a case study in which we apply REANN to neural networks used in a more complex context: time series prediction. We devise some modifications to REANN to adapt this algorithm to the problem of time series prediction. Experimental results on three real world time series datasets demonstrate the effectiveness of the proposed approach in generating accurate rules from neural networks for time series prediction.