Parallel reducts for incremental data

Parallel reducts are more suitable for dynamic data, incremental data, and multi-source data than other reducts, and can be obtained by attribute significance in a family of decision subsystems. However, when data are increasing, they should be improved or changed to fit the new dataset. In this paper, some properties of parallel reducts for changing data are discussed, and an algorithm for improving parallel reducts is proposed. Some improved ideas of the algorithm are introduced to fit decreasing data and changing data. Experimental results show that the algorithm can reduce most of time for calculating a new parallel reduct when new data are increasing.