Impact of Cover Parameter Value on Rule Generation in Rough Set Classification

Machine learning plays a crucial role in healthcare classification, with Rough Set Theory (RST) offering effective tools for managing data uncertainty. Within RST, the RSES2 tool supports algorithms like LEM2 and Covering, yet the influence of cover parameter values on rule generalization and specificity remains underexplored. This study investigates these effects using the Differentiated Thyroid Cancer dataset. The research investigates the trade-offs between rule generalization and specificity by adjusting cover parameter settings, which dictate the minimum and maximum cases a rule must cover. The comparison reveals that the LEM2 algorithm maintains high accuracy across various cover parameter values, with only a slight decline as the parameter increases, and shows improved coverage with higher cover values. In contrast, the Covering algorithm displays greater fluctuations in accuracy, peaking at lower cover parameter values and decreasing significantly as the parameter rises. Coverage for the Covering algorithm is highest at lower cover parameters but decreases sharply at higher values. This indicates that LEM2 is more robust in maintaining accuracy and coverage, while the Covering algorithm performs better at lower cover parameters but struggles with stability as the parameter increases.