Supervised Machine Learning Approaches for Multimodal Soil and Plant Health Monitoring in Precision Agriculture

Abstract

Precision agriculture increasingly depends on intelligent data-driven systems to enhance crop productivity while supporting sustainable resource management. Early and accurate identification of plant stress is essential for timely intervention and improved yield. This study presents a supervised machine learning framework for real-time soil and plant health monitoring using multimodal sensor data. The proposed approach integrates environmental factors, soil parameters, and plant physiological measurements to provide a comprehensive assessment of crop health conditions. A heterogeneous dataset of sensor readings is utilized to classify plant health into three categories: Healthy, Moderate Stress, and High Stress. Four supervised machine learning algorithms, K-Nearest Neighbors (KNN), Decision Tree, Random Forest, and Gradient Boosting, are implemented, optimized, and evaluated using standard performance metrics, including accuracy and the Area Under the Receiver Operating Characteristic Curve (AUC). The evaluation strategy ensures reliable comparison across models. Experimental results demonstrate that ensemble-based models significantly outperform non-ensemble approaches in capturing the complex and nonlinear relationships present in agricultural sensor data. Random Forest and Gradient Boosting achieve near-perfect classification performance, with the Random Forest model attaining an AUC score of 1.00, indicating excellent class separability. The Decision Tree model also shows strong predictive capability, while the KNN model records comparatively lower accuracy of approximately 64%, reflecting its limited suitability for this application. The findings confirm the effectiveness of ensemble machine learning techniques as robust decision-support tools for precision agriculture and establish a practical baseline for model selection in real-time soil and plant health monitoring systems.