Modeling Users Behavior in Cellular Networks with Mobile Big Data

Studying the behavior characteristics of mobile communication users may provide good guidance in resource management, network optimization and etc. Therefore, we analyze the call detail record (CDR) data by fitting, and find several laws of users behavior. First of all, we found nonhomogeneous Poisson process is more suitable to model call arrival process than homogeneous Poisson process because of the unbalanced traffic which caused by people's living patterns. Meanwhile, we applied the compound Poisson process to fit the call duration data. Secondly, we analyzed the number of different base stations that users access, and found it conforms to the log-normal distribution. Moreover, truncated exponential log-normal (TELN) model was proposed to describe the number of calls users have made monthly. Finally, Gaussian mixture model (GMM) was adopted to describe the relationship between call arrival and time.