Improvements to Data Reconstruction in IoT Sensor Networks Under Realistic Conditions

IoT sensor networks are gaining in popularity with their low cost sensors and inexpensive, albeit opportunistic, management. However, their data reconstruction processes are lagging behind, often resorting to a simple averaging or interpolation. This paper examines the feasibility of directly applying artificial neural networks (ANNs), and more specifically the multi-layer perceptron (MLP), to improve the quality of the data reconstruction in IoT sensor networks, taking into account realistic operating conditions. Four cases are analyzed with the use of a simulation: normal operation, sensor failure, noise and a long-term drift, then compared with the use of interpolation. The results indicates that, in all those situations, MLP can indeed deliver improvements to data reconstruction.