Machine learning-driven IoT framework for soil moisture prediction in precision irrigation systems

Abstract The agricultural sector faces a significant challenge in increasing food production for a growing population under conditions of limited water resources. This paper proposes an innovative Internet of Things and machine-learning (ML) enabled system for predicting and monitoring soil moisture (SM) content using sensors. The proposed system uses a SM sensor, a digital thermometer22 sensor (temperature and humidity), a DS18B20 sensor, a light intensity sensor, and a pressure sensor to measure various characteristics of the environment in real time. The research involved three types of soils (loam, sand, and silt) from which real-time data were collected and used to calibrate the sensors to improve the precision of the ML model, thereby allowing accurate prediction of SM. Statistical analysis was performed on the dataset to understand non-linearity and feature dependency among soil parameters. ML algorithms (Random Forest, extra trees, Xtreme gradient boosting, LightGBM and CatBoost were developed and validated for the three types of soil. Further, we proposed a StackingSM model and validated it using a 5-fold cross-validation approach for robustness and generalizability across different data subsets. The proposed StackingSM model achieved highest ( R 2 ) accuracy on loam soil (0.96), followed by sandy soil with an R 2 of 0.93 and silt soil with an R 2 of 0.90. The proposed framework supports data-driven decision-making for precision irrigation by providing accurate SM predictions under varying soil conditions.