Automatic determination about precision parameter value based on inclusion degree with variable precision rough set model

The rough set theory provides a powerful approach for attributes reduction and data analysis. The variable precision rough set (VPRS) model, an extension of the original rough set approach, tolerates misclassifications of the training data to some degree, which promotes the applications of rough set theory in inconsistent information systems. However, in most existing algorithms of feature reduction based on VPRS, the precision parameter (β) is introduced as prior knowledge, which restricts their applications because it is not clear how to set the β value. By studying β-consistency in the measurement of a decision table and the threshold value of the β-consistent decision table, this paper presents an algorithm for automatic determination of the precision parameter value from a decision table based on VPRS. At the same time, the precision parameter value from our proposed method is compared with the thresholds from the decision-theoretic rough set (DTRS). The influences of the precision parameter are also discussed on attribute reduction, which shows the necessity of the estimated precision parameter from a decision table. The simulation results including VPRS and other classification methods in real data further indicate that different precision parameter values make a great difference on rules and setting a precise parameter near the threshold value of the β-consistent decision table can precisely reflect the decision distribution of the decision table.