A new rough neural network (RNN)-based model is proposed in this paper. The radial basis function network with dynamic decay adjustment (RBFNDDA) is applied to learn information directly from a data set and group it in terms of prototypes. Then, a neighborhood rough set-based procedure is applied to detect prototype outliers. This hybrid model is named rough RBFNDDA1. However, the removal of all outliers may cause information loss because some outliers may represent rare yet useful information in a classification task. As such, the parameters of a prototype outlier, i.e., its radius and weight, are exploited to gauge whether the information encoded by the prototype is meaningful. This hybrid model is named rough RBFNDDA2. The results from a benchmark experimental study show that rough RBFNDDA2 can retain meaningful prototype outliers and, at the same time, significantly reduce the number of prototypes from the original RBFNDDA model while maintaining classification accuracy. A real-world application in a power generation plant is used to evaluate and demonstrate the effectiveness of the proposed model. [ABSTRACT FROM AUTHOR]