Structural Analysis of User Association Patterns in Wireless LAN

by Wei-jen Hsu, Debojyoti Dutta, & Ahmed Helmy arXiv.org E-print Archive, 1 Jun 2006 Due to the rapid growth in wireless local area networks (WLANs), it has become important to characterize the fine-grained structure of user association patterns. We focus on unraveling the structure in user's daily association patterns in WLANs in the long run. The daily access pattern is defined by the fraction of time it spends with a particular location. We answer three questions: 1) Do users demonstrate consistent behavior? Using our novel metrics and clustering, we conclude that many users are multi-modal. 2) Is it possible to represent user association patterns using a compact representation? Using eigen-decomposition, we show that the intrinsic dimensionality of the constructed user association matrices is low and only the top five eigenvalues and their corresponding eigenvectors can be used to reconstruct those association matrices with an error of 5%, in terms of the L1 and L2 matrix norms. 3) How can we decide if two users have similar association patterns? We define two new metrics, and we rigorously validate their efficacy by demonstrating that the inter and intra cluster distributions, upon clustering, have very little overlap. Our methods and observations are a first step towards systematically mining user-association patterns and could lead to new directions in network management and understanding social patterns of users. Read more