Indoor Temperature Prediction in an IoT Scenario.

One of the hottest topics being researched in the field of IoT relates to making connected devices smarter, by locally computing relevant information and integrating data coming from other sensors through a local network. Such works are still in their early stages either by lack of access to data or, on the other hand, by the lack of simple test cases with a clear added value. This contribution aims at shading some light on how knowledge can be obtained, using a simple use case. It focuses on the feasibility of having a home refrigerator performing temperature forecasts, using information provided by both internal and external sensors. The problem is reviewed for both its potential applications and to compare the use of different algorithms, from simple linear correlations to ARIMA models. We analyse the precision and computational cost using real data from a refrigerator. Results indicate that small average errors, down to ≈0.09 ∘ C, can be obtained. Lastly, it is devised how can the scenario be improved, and, most importantly, how this work can be extended in the future.