Composability of Markov Models for Processing Sensor Data

We show that it is possible to apply the 'divide-and-conquer' principle in constructing a Markov model for sensor data from available sensor logs. The state space can be partitioned into clusters, for which the required tran- sition counts or probabilities can be acquired locally. The combination of these local parameters into a global model takes the form of a system of linear equations with a confined solution space. Expected advantages of this approach lie for example in reduced (wireless) communication costs.