摘要：This paper addresses the problem of data aggregation platforms operating in heterogeneous Ambient Intelligence Environments. In these platforms, device interoperability is a challenge and erratic sensor observations are difficult to be detected. We propose ADES (Automatic Detection of Erratic Sensors), a statistical approach to detect erratic behavior in sensors and annotate those errors in a semantic platform. To do that, we propose three binary classification systems based on statistical tests for erratic observation detection, and we validate our approach by verifying whether ADES is able to classify sensors by its observations correctly. Results show that the first two classifiers (constant and random observations) had good accuracy rates, and they were able to classify most of the samples. In addition, all of the classifiers obtained a very low false positive rate.
