## 8.2 PREDICTIVE LOCATION TRACKING TECHNIQUES

Owing to the uncertainty inherent in the mobile's movements, we can consider them to be the outcome of an underlying stochastic process, which can be modeled using established information-theoretic concepts and tools [34, 56]. The cornerstone work of Ref. [17] exhibited the possibility of using methods, which have traditionally been used for data compression (thus, characterized as “information-theoretic”), in carrying out prediction. Considering a symbolic network topology model [8], we can model the respective state space as a finite alphabet comprised of discrete symbols. The alphabet consists of all possible sites (cells) where the client has ever visited or might visit (assuming that the number of cells in the coverage area is finite). With this transformation, we can exploit methods that have traditionally been used for data compression (thus, characterized as “information-theoretic”) to carry out prediction. In the rest of this section, we elaborate on these methods.

### 8.2.1 The Discrete Sequence Prediction Problem

In quantifying the utility of the past in predicting the future, a formal definition of the problem is needed, which we provide in the following lines. Let Σ be an alphabet, consisting of a finite number of symbols *s*_{1}, *s*_{2}, … , *s*_{|Σ|}, where | · | stands for the length/cardinality of its argument. A *predictor*, which is an algorithm used to generate prediction models, accumulates sequences of the type , where , ∀*i*, *j* and ...