Trust Model for the Internet of Things

Resumen

The number of devices used in multiple fields is increasing more and more, as well as the amount of data they generate on the Internet of Things. But the datasets produced by sensors may have errors produced because of multiple issues (such as faulty devices or vandalization of sensors). Therefore, it is necessary to use mechanisms that can determine whether there are problems with the sensors. This work proposes to analyse the behaviour of a sensor based on the data it has generated, looking at two main aspects: how the data produced varies and to what extent the data contains faulty values that might be problematic. The proposed approach is focused mainly on the combination of some solutions based on statistics, in such a way that the outcome will be as much generic as possible. This dissertation introduces all the heterogeneous data sources that have been used (and their particularities) and it performs a deep analysis of several statistical aspects for many types of sensors. It addresses the probability distribution of the data, as well as some statistical tests for studying if the data varies like white noise. Then, it looks at the statistical tests that analyse the presence of outliers and homogeneity, as they can be linked to certain types of errors in sensors. All these results are combined, defining three models that analyse the variation of the data, the presence of outliers and the potential correlation with other sensors, giving as a result the implementation of a mechanism for sensors¿ data understanding and trust evaluation. The evaluation of the solution has shown a good performance with datasets and types of sensors not studied before, with an accuracy of around 0.8 and F0.5-score higher than 0.7 in most of the cases.