The SWAMP project published a paper in the 2020 IEEE International Workshop on Metrology for Agriculture and Forestry (MetroAgriFor 2020) reporting a data filtering mechanism for optimizing data transmission between fog and cloud in an IoT system. A Nearest Neighbors based Data Filter for Fog Computing in IoT Smart Agriculture proposes an approach to collect and store data in a smart agriculture environment and two different methods filtering data in the fog. We designed an experiment for each filtering method, using a real dataset containing temperature and humidity values. In both experiments, the fog filters the data using the k-Nearest-Neighbors (kNN) algorithm, which classifies data into categories according to their value ranges.
The dataflow is presented in the picture below. Sensors collect data from a farm environment and send it to the fog. The data arrives in the fog through a network server, transferring the raw data to the data storage module and the data analytics module. The data analytics module analyses the raw data to make decisions, then, it stores the decisions in fog memory and sends the decisions to actuators to irrigate a crop. The data storage module is responsible for storing data in memory. It can provide past decisions and filtered data to the analytics module. Furthermore, the data storage module provides the raw data periodically to the fog data filter.
We divided the classification process into two rounds: for the round 1, we classified the data collected between 00:00 and 06:00 and for the round 2, between 06:01 and 12:00 (both on the first day of August 2013). The chart below shows a correlogram for each round comprising four graphic plots. The correlogram includes two density plots and two dispersion plots, highlighting the category density according to temperature and humidity. In some density curves, we observed that classes overlap, although it does not prevent data classification from occurring because we deal with two attributes
