Feature subset selection for multi-scale neighborhood decision information system via mutual information

As a granular computing model, multi-scale data analysis has attracted considerable attention in last several years. However, most of multi-scale models are hardly to deal with multi-source data, especially in heterogeneous environments. For this reason, we investigate a novel multi-scale model by combining weighted neighborhood rough sets with Wu-Leung model for the first time, and apply it in feature selection for multi-source heterogeneous multi-scale data. First, the multi-scale weighted neighborhood granules obtained and their properties are discussed. Second, the mutual information of multi-source heterogeneous multi-scale features is presented. Based on this, the definition of the redundancy of the features is obtained and a feature subset selection algorithm that simultaneously performs the selection of features and the optimal scale combination is given. Finally, numerical experiments on multi-source heterogeneous multi-scale datasets and heterogeneous multi-scale datasets are conducted to examine the effectiveness and feasibility of the proposed model. The experiments demonstrate that the proposed model can obtain better results on both datasets.