In this paper, we address the problem to extract loyal visitors' characteristics at central commercial districts by using visit frequency data from on-site sampling survey. These kinds of data normally collected by home-based sampling survey because the phenomenon of consumer's trip occurs at their home. And yet it takes expensive cost, so we usually adopt on-site sampling survey. But if we apply the data collected by on-site sampling survey to the methods of statistical estimation, it is known that choice-based bias problem is caused. The purpose of this paper is to propose the modified regression tree method, which corrects the choice-based bias and to apply this method to actual data of Fukuoka City in Japan. JEL classification: M31, R58