Confidence Estimation of the State Predictor Method

Pervasive resp. ubiquitous systems use context information to adapt appliance behavior to human needs. Even more convenience is reached if the appliance foresees the user’s desires. By means of context prediction systems get ready for future human activities and can act proactively. Predictions, however, are never 100% correct. In case of unreliable prediction results it is sometimes better to make no prediction instead of a wrong prediction. In this paper we propose three confidence estimation methods and apply them to our State Predictor Method. The confidence of a prediction is computed dynamically and predictions may only be done if the confidence exceeds a given barrier. Our evaluations are based on the Augsburg Indoor Location Tracking Benchmarks and show that the prediction accuracy with confidence estimation may rise by the factor 1.95 over the prediction method without confidence estimation. With confidence estimation a prediction accuracy is reached up to 90%.