Abstract: The original reduction set may be invalid when new data are added to the decision system.Most of the existing incremental algorithms focus on increasing attributes or increasing samples.This paper analyzes the changing rules of the decision system after adding new attributes and samples.An improved incremental attribute reduction algorithm is presented with the framework of neighborhood rough sets,which can update the original reduction set dynamically by using the idea of relative positive region,and handle the two situations of incremental data mentioned above at the same time.The time complexity of the presented algorithm was analyzed and compared with the classical algorithm.The experiment results show that this conclusion accords with the attribute reduction obtained from traditional algorithm and the efficiency is improved.

Key words: neighborhood system, incremental learning, relative positive region, attribute reduction