A novel ANN-based feature subset selection in multi-scale granular ball neighborhood decision tables

As an effective data preprocessing method, feature subset selection has been widely explored in recent years. However, the feature subset selection for the Wu-Leung model and its extended model involves high time complexity. Therefore, we combine the granular ball neighborhood rough set with the Wu-Leung model. A multi-scale granular ball neighborhood decision table is designed. Meanwhile, the weight of features are not the same at different scales and are correlated with the decision. We combine the artificial neural network (ANN) with the multi-scale granular ball neighborhood decision table to calculate feature weights. Based on this, the weight-based feature subset selection algorithm, and the positive region increment-based feature subset selection algorithm are designed. Most importantly, in order to overcome the situation of missing or missing labels during data collection, we design an unsupervised feature subset selection algorithm. Finally, the effectiveness and feasibility of the proposed three algorithms are verified through experimental comparison and analysis. The advantages of the proposed algorithm are analyzed.