The concentration of heavy metals that needs to be maintained in aquaponic environments for habitable growth of plants has been a cause of concern for many decades now as it is not possible to eliminate them completely in a commercial set-up. Our goal is to design a cost-effective real-time smart sensing and actuation system in order to control the concentration of heavy metals in aquaponic solutions. Our solution consists of sensing the nutrient concentrations in the aquaponic solution, namely calcium, sulfate and phosphate and providing them to the Machine Learning model hosted on an Android application. The application outputs the appropriate iron and copper concentrations that can be tolerated for optimal growth of plants in an aquaponic set-up and controls the dispensing systems in order to maintain these desired heavy metal concentrations.  

The Machine Learning algorithm used in this case is pre-trained on the top three nutrient predictors selected from a dataset containing the nutrient profiling of samples recorded from three aquaponic farms over the course of a year in South-East Texas based on the output of a pipeline of Feature Selection models like the pairwise correlation matrix, ExtraTreesClassifier and Xgboost classifier. This pre-trained ML classification model, which in our case is a Radial Support Vector Machine, is hosted on a cloud platform and would output the recommended levels of iron and copper in real time through an Android application considering the concentrations of phosphorus, calcium and sulfur as inputs. These recommended values were maintained with the help of an array of dispensing and sensing units, thus monitoring these parameters in a closed loop system.