SoilFlux: An Embedded IoT-Machine Learning Framework for Sustainable Agriculture

Abstract

Sustainable agriculture is hindered by high cost and low availability of conventional physical sensors for nitrogen, phosphorus and potassium (NPK) that prevents smallholder farmers from using data-driven agronomic decision-making. This investigation proposes a cost-effective Internet-of-Things framework that virtually predicts NPK concentrations, which helps overcome the technological shortfall that has been identified. The proposed system involves coupling an ESP32 microcontroller with DHT11 and soil moisture sensors to obtain proxies for the environment which are processed using Random Forest regression model. Empirical evaluation shows that the Random Forest algorithm achieves the highest performance of prediction with an R2 score of 0.902. Moreover, the fusion of SHAP and LIME analyses gives the model interpretability and hence legitimizes the system as a reliable decision supporting tool consistent with Sustainable Development Goal-2 (Zero Hunger), Goal-6 (Clean Water and Sanitation), and Goal-12 (Responsible Consumption and Production).