摘要：In the field of data mining, spatial drift refers to the data used to develop the model consists of only one part of the population, and the differences among the samples or between sample and population are unknown. This paper proposes a two-step Bayes method to improve adaptability for different region samples, which also maintains high model accuracy. The new method first groups region based on similarity, second, sets a model structure without parameters for populations or large samples with good data quality, and then trains parameters using samples in same region group. This method builds a estimation model, proving the method by showing how it can to some extent solve the uncertainty of consumer classification
