Towards Precision Agriculture: A Real-Time Soil Fertility Monitoring System Using IoT and Deep Learning

Soil fertility is a critical determinant of agricultural productivity and sustainability. This study presents a deep learning framework for classifying soil fertility levels based on a comprehensive dataset of 880 soil samples, each characterized by 12 parameters: Nitrogen (N), Phosphorus (P), Potassium (K), pH, Electrical Conductivity (EC), Organic Carbon (OC), Sulfur (S), Zinc (Zn), Iron (Fe), Copper (Cu), Manganese (Mn), and Boron (B). The data underwent rigorous preprocessing, including handling missing values, removing duplicates, addressing class imbalance via oversampling, and eliminating outliers using the Interquartile Range (IQR) method. An Artificial Neural Network (ANN) model was developed for multi-class classification, featuring three hidden layers with ReLU activation and HeNormal initialization, and a softmax output layer. The model, trained with the Adam optimizer and sparse categorical cross-entropy loss, achieved a high validation accuracy of 93.04% with a loss of 0.2102. Furthermore, this research integrates an IoT-based system utilizing sensors such as the DS18B20 for temperature and NPK sensors for nutrient monitoring to enable real-time soil condition assessment. The synergy of machine learning and IoT technologies established in this work provides a scalable, e fficient framework for precision agriculture, with the potential to enhance crop yield, optimize resource use, and promote sustainable farming practices.