The connectionist approach to multivariables forecasting of precipitation with virtual term generation schemes

Time series prediction is the prediction of future measurements by analyzing the relation among past values and a current observation. Many papers propose the neural approach to this instead of statistical approaches because neural network outperforms the statistical methods in time series prediction. If the neural approach replaces the statistical ones, it requires sufficient data for training. This paper proposes the schemes to generate artificially more data by estimating X(t+0.5), based on interpolation. The data for the experiments in this paper is about the precipitation of the three areas, east, middle, and west, in State Tennessee of the USA. The prediction performance is improved by more than 60% using the virtual term generation.