Prediction of Indoor Movements Using Bayesian Networks

Abstract. This paper investigates the efficiency of in-door next location prediction by comparing several prediction methods. The scenario concerns people in an office building visiting offices in a regular fashion over some period of time. We model the scenario by a dynamic Bayesian network and evaluate accuracy of next room prediction and of duration of stay, training and retraining performance, as well as memory and performance requirements of a Bayesian network predictor. The results are compared with further context predictor approaches- a state predictor and a multi-layer perceptron predictor using exactly the same evaluation set-up and benchmarks. The publicly available Augsburg Indoor Location Tracking Benchmarks are applied as predictor loads. Our results show that the Bayesian network predictor reaches a next location prediction accuracy of up to 90 % and a duration prediction accuracy of up to 87 % with variations depending on the person and specific predictor set-up. The Bayesian network predictor performs in the same accuracy range as the neural network and the state predictor.