In this paper, the concept of Restricted Dissimilarity Functions on \(L^n\) (RDFn) is introduced to address limitations in comparing multi-attribute data with RDFs, for example, for comparing, maintaining, and even distinguishing the relevance of individual attributes. A construction method for RDFn is provided and applied to the proposed k-Nearest Neighbors model, denoted by kNN-RDFn, for short-term precipitation image classification is proposed. Results of extensive experiments show that kNN-RDFn significantly improves classification performance over traditional models.
