Research on Complete Algorithms for Minimal Attribute Reduction

Minimal attribute reduction plays an important role in both theory and practice, but it has been proved that finding a minimal reduct of a given decision table is a NP-hard problem. Some scholars have also pointed out that current heuristic algorithms are incomplete for minimal attribute reduction. Based on the decomposition principles of a discernibility function, a complete algorithm CAMARDF for finding a minimal reduct is put forward in this paper. Since it depends on logical reasoning, it can be applied for all decision tables after their discernibility functions constructed reasonably. The efficiency of CAMARDF is illustrated by experiments with UCI data sets further.