• 목록

The patient rule induction method (PRIM) is a data-driven approach for response optimization that identifies desirable operating conditions as axis-aligned subregions of the input space. When stochasticity is incorporated into PRIM, numerous high-performing subregions can be generated from operational data; however, interpreting their structural relationships and identifying meaningful groups remains challenging. This study proposes a granularity-based heuristic for organizing stochastic PRIM subregions into effective families of operating conditions. The proposed framework measures empirical overlap among subregions, applies hierarchical clustering, and evaluates candidate cluster structures using effective cluster compactness while maintaining sufficient coverage of the discovered subregions. A case study using real-world operational data demonstrates that the method can identify internally compact clusters that are repeatedly rediscovered across stochastic PRIM trials. These clusters provide interpretable summaries of recurring high-performing operating regions and can support process engineers in selecting stable and practically meaningful operating conditions for data-driven process control.