In this chapter, we have provided an overview of different approaches to predicting the location of users in a mobile wireless system. This chapter is not intended to be a comprehensive survey, and in particular we have only summarized a few of the approaches being used in domain-specific algorithms for location prediction.
We see two general ideas pursued in the literature: domain-independent algorithms that take results from Markov analysis or text compression algorithms and apply them to prediction, and domain-specific algorithms that consider the geometry of user motion as well as the semantics of the symbols in the movement history. Domain-independent algorithms tend to have well-founded theoretical principles on which they are based and can make analytical statements about their prediction accuracy. However, in some cases, these statements refer to the asymptotic optimality of their accuracy, i.e., that as the input history approaches infinite length, no similar prediction algorithm can do any better. While satisfying from a theoretical point of view, it is unclear how relevant these results are in practice. On the other hand, some domain-specific algorithms offer heuristics that appear intuitively appealing but have no explicit theoretical analysis to support them. Clearly, a better bridge between engineering intuition and theoretical analysis would be helpful.
One of the major barriers to practical advancement in this area is the lack of publicly available empirical data to guide future research. Most studies have used artificial mobility models; relatively few, e.g., Chan and coworkers,  have collected empirical data for the domain of interest (cellular handoffs) and used them for validation. We compare the situation to the early work done on caching disk pages in computer systems. A large variety of cache replacement policies, many of which were intuitively plausible, were proposed. It was only empirical data from page fault traces that enabled the conclusion that the Least Recently Used (LRU) algorithm offered the best compromise between simplicity and effectiveness in most cases. Large-scale statistical data for the domains of interest is sorely needed to help provide benchmarks and directions for future research.
Chan, J., Zhou, S., and Seneviratne, A., A QoS adaptive mobility prediction scheme for wireless networks, Proc. IEEE Globecom, Sydney, Australia, Nov. 1998.