This paper introduces an approach of Weighted Data Normalization (WDN) for Radial Basis Function (RBF) type of neural networks. It presents also applications for medical decision support systems. The WDN method optimizes the data normalization ranges for the input variables of the neural network. A steepest descent algorithm (BP) is used for the WDN-RBF learning. The derived weights have the meaning of feature importance and can be used to select a minimum set of variables (features) that can optimize the performance of the RBF network model. The WDN-RBF is illustrated on two case study prediction/identification problems. The first one is prediction of the Mackey-Glass time series and the second one is a real medical decision support problem of estimating the level of renal functions in patients. The method can be applied to other distance-based, prototype learning neural network models.