In order to forecast the short-term customer flow in trading area under wireless access, through analyzing of customer shopping behaviors, this paper presents an identification scheme based on staying time and activeness to distinguish the staff and customers. We use the binary linear regression method to fit the data under confidence level of 95%, and analyze the influence of different parameters to predict. Experimental results show that the staying time and activeness are reasonable and effective to distinguish the identity information, when time threshold is 3 and activeness threshold is 2, the wavelet neural network prediction effect is best.