Optimal scale combination selection based on a monotonic variable precision multi-scale rough set model

Most existing generalized multi-scale rough set models (GMRSMs) are based on Pawlak's rough set, which lacks fault tolerance and thus limits their generalization ability. To improve generalization, the variable precision generalized multi-scale rough set model (VPGMRSM) was proposed. However, this model disrupts the monotonicity of the positive region, posing challenges for optimal scale combination (OSC) selection. To address these issues, a monotonic VPGMRSM is proposed in this paper through a two-stage approximation process. The proposed model preserves the monotonicity of the GMRSM and the fault tolerance of the VPGMRSM, and is further applied to OSC selection. First, the non-monotonicity of the positive region in the original VPGMRSM is analyzed. Then, a monotonic VPGMRSM is proposed, whose information measurements are proven to satisfy the monotonicity lacking in the original model. Second, an extended definition of OSC is proposed based on the positive region in the new model, which significantly simplifies and improves the efficiency of the OSC selection process. Third, two OSC selection algorithms are proposed: one based on binary search to find a single OSC, and the other based on three-way decision theory to identify all OSCs. Finally, the experimental results validate the monotonicity of the positive region in the new model and demonstrate that the proposed algorithms are not only suitable for VPGMRSMs, but also effectively reduce the computation time.