Abstract: The original RFID data stream contains a lot of noise and uncertainty,so the data must be cleaned before using and the cleaning strategy is the guarantee of the quality of the cleaning.In this paper,a new method for cleaning the RFID data stream was proposed.The maximum entropy principle is introduced in the cleaning strategy,and this treat cleaning RFID tuple attributes to select weights.the cleaning cost analysis is performed according to the cleaning node time-consuming and error to decide the best cleaning method.Simulation experiment results show that this clea-ning strategy improves the cleaning efficiency and accuracy of the RFID data stream.

Key words: RFID data stream,Uncertainty,Cleaning strategy,Cleaning costs,Max-entropy feature selection