Due to the further spread of the broadband, the network of broadband access points becomes quite complicated. It becomes a serious problem that how to define and analyze the behavioral features of large amount of access points. The features of access points are studied based on the log records of users’ Internet account. Several useful features are defined and calculated, and then machine learning is introduced to classify the broadband access points. The verification of experimental results shows that it can be identified whether a broadband access point falls into residential category or non-residential category and its behavioral features can be described accurately and efficiently.