Improving Next Location Prediction by Using Hybrid Predictors

Abstract. Neural networks, Bayesian networks, Markov models, and state predictors are different methods to predict the next context in a ubiquitous environment. For all methods a lot of parameters must be set up which differ for each user. Therefore a complex configuration must be made before a prediction method can be used. A hybrid predictor can reduce the configuration overhead. Such a hybrid predictor uses different prediction methods or configurations in parallel yielding different predic-tion results. A selector selects a prediction result from the result set of the base predictors. We propose and evaluate three principal hybrid pre-dictor approaches – the warm-up predictor, the majority predictor, and the confidence predictor – with several variants. The hybrid predictors reached a higher prediction accuracy than the average of the prediction accuracies of the separately used prediction methods.