A Prediction-Based Multisensor Heuristic for the Internet of Things

As most of IoT devices produce large volume of data in low semantic levels and are limited on resources such as memory and processing capacity, the usage of traditional machine learning techniques on resource constrained devices to support real-time decision making are unfeasible. In this paper, we propose a heuristic for adapting data prediction and data fusion techniques to preprocess data and avoid unnecessary communication between sensor devices and the Sink node, which would infer the actual reading within an predefined margin, deciding when sensors send, or Sink request data. In this work we compared (i) linear estimation; (ii) Weightless Neural Networks; and (iii) Moving Average Convergence Divergence. As main contribution, our proposed algorithms show how to mitigate unnecessary communication when supported by data fusion and data prediction techniques.