Mobile Based IoT System for Real -Time Soil Monitoring and Machine Learning Predictive Irrigation in Smart Agriculture

Abstract

Water scarcity and inefficient irrigation remain problems in agriculture. This study developed and tested an IoT-based irrigation system with a mobile application and machine learning prediction for rice paddy water management. Soil moisture, air temperature, and relative humidity sensors were connected to an ESP32 microcontroller, which collected data and sent them to a cloud platform. A total of 8,640 valid observations were collected over 30 days from two plots at 10-minute intervals. Four algorithms (Decision Tree, Random Forest, Support Vector Machine, and Gradient Boosting) were trained and compared to predict irrigation need from current and historical conditions. The mobile application showed real-time readings, irrigation predictions, and historical data, and allowed manual pump control. The system reached a data transmission success rate of 98.7% and an average response time of 3.8 s. Random Forest gave the highest accuracy (94.3%), with a precision of 92.1%, a recall of 90.8%, and an F1-score of 91.4%. Compared with threshold-based irrigation, the predictive strategy used 20.8% less water (1,016.8 L vs. 1,284.6 L) and 18.4% fewer irrigation events (151 vs. 185), and soil moisture varied less (SD 5.7% vs. 7.4%). These results come from a 30-day trial with two plots at one site and were not tested statistically. Future work includes multi-depth soil sensing, weather forecast integration, solar-powered operation, and multi-season field validation.