A method of selecting learning data in the prediction of time series with explanatory variables using neural networks

In the prediction of time series using multilayer feedforward neural networks, there are two practical methods for selecting learning data: the moving window data learning method, and the similar data selective learning method with the correlation coefficient based similar data selection method which we proposed in a previous paper. In this paper, for time series data with explanatory variables, the predictive performance by the two methods was investigated by numerical simulations. With the time series whose nature is choppy, the latter performed considerably better than the former. With the time series whose nature is smooth, the former performed slightly better than the latter. According to these results, it was found that the latter is effective for a time series whose nature is choppy