文摘
Wireless and in particular 802.11 is one of the major technologies for accessing the Internet at home, in coffee shops, enterprises, university campuses, and other public places. While most recent works on modeling wireless sites focuses on user mobility and user residing time, this paper presents and compares a number of models for characterizing access point (AP) usage including time-dependent models that considers week structure usage. Moreover, rather than looking at throughput we focus on daily counts of keep-alive events that mobile devices generate every 15?min while they are connected to the wireless network. We model both daily event counts and above–below AP event counts average binary indicator. Our models are trained and evaluated on data collected from Porto hotspot of Eduroam, the European academic wireless network. The models we present are generative, in the sense they can be used to generate synthetic daily event counts for a single AP or a collection of APs. We provide standard cross-validation comparison of models using the log-likelihood of the models on training and test data. We conclude that significant improvements in AP usage modeling capability can be observed by considering (1) simple time dependency (2) week-days/week-ends usage structure and (3) individual day’s usage; whereas extending the complexity of time dependency ordering of AP’s usage samples does not show significant improvements for daily event count models.