IoT-Integrated Machine Learning for Real-Time Soil Nutrient Monitoring and Horticultural Crop Recommendations

Farmers and agricultural newcomers often rely on intuition for plant nutrient management, leading to inaccuracies. This study presents a real-time soil nutrient monitoring and plant recommendation system using IoT and machine learning. The prototype integrates an ESP32 microcontroller, NPK sensor, soil pH sensor, and DHT11 with an application and a machine learning model. The model employs a hard voting method combining random forest, LightGBM, and support vector machine algorithms, deployed via Flask API and hosted on Vercel. Testing showed the system effectively monitors soil nutrients in real-time, with a pH sensor error margin of ±0.51 and NPK sensor readings displaying a non-linear increase in levels. The hard voting method achieved 98.5% accuracy, with recommendation processing taking less than one second (371.17ms) and CPU usage at 7%. System performance depends on internet speed, while recommendation accuracy is influenced by temperature and humidity. This system offers a practical solution for precision agriculture by optimizing nutrient application and plant recommendations.