Accelerating incremental attribute reduction algorithm by compacting a decision table

Wei Wei, Peng Song, Jiye Liang, Xiaoying Wu

Abstract: The evolution of object sets over time is ubiquitous in dynamic data. To acquire reducts for this type of data, researchers have proposed many incremental attribute reduction algorithms based on discernibility matrices. Although all reducts of an updated decision table can be obtained using these algorithms, their high computation time is a critical issue. To address this issue, we first construct three new types of discernibility matrices by compacting a decision table to eliminate redundant entries in the discernibility matrices of the original decision table. We then demonstrate that the set of reducts obtained from the compacted decision table are the same as those acquired from the original decision table. Extensive experiments have demonstrated that an incremental attribute reduction algorithm based on a compacted decision table can significantly accelerate attribute reduction for dynamic data with changing object sets while the acquired reducts are identical to those obtained using existing algorithms.