A Fuzzy Neural Network Precipitation Model Established by Blurring the Rough Set Factors

To improve the predictive ability of a fuzzy neural network prediction model, the re-selection is made, by means the rough set attribute reduction, of the correlated prognostic factors that have been chosen and the re-selected factors are treated by blurring as model input, thereby establishing a new-type fuzzy neural network predictive model. Experiments are conducted for approximately two months with day-to-day mean rainfall as the predictive target. Result shows that the presented model that results from a, new technique for choosing prognostic factors and a processing scheme is superior to the conventional regression and fuzzy neural network prediction models, leading to appreciably higher precision of results compared to the latter two. Eventually, the merits of the rough set attribute reduction and blurring techniques are explained.