Application of Bayesian Belief Networks for context extraction from wireless sensors data

Networks of Wirelessly connected low power sensors have ability to closely sense activity of individual and social interest. The usefulness of Wireless Sensor Networks is increased further by deriving contextual information from it. From sensors data, context like activity, location, weather and surroundings (nearby persons, devices) can be deduced. Techniques to represent & extract the context include ontology, Markov Models, decision trees, clustering and Bayesian approaches. Given incomplete and erroneous nature of sensor data, Bayesian Belief Networks (BBN) are used here to obtain features defining context. Five algorithms of BBN construction have been evaluated for comparing feature classification performance. Simple rule based matching is then applied to map the features to already defined context. The mechanism is applied here on sensors data obtained from Intel research lab at Berkeley to extract the “weather” context. Similar mechanism can be applied to other application and contexts also.