Comparative analysis of machine learning models for smart irrigation systems

Abstract

Intelligent irrigation systems play a crucial role in addressing the global issues of water scarcity, climate variability, and sustainable agricultural production. These systems can help identify the efficient time and the exact quantity of irrigation through the use of data-driven ideas, which ensures maximum crop yield with minimal use of water. This paper provides a thorough comparative analysis of the four most commonly used Machine Learning (ML) models: Support Vector Machines (SVM), Gradient Boosting (GB), K-Nearest Neighbors (KNN), and Logistic Regression (LR), to predict the need of irrigation based on critical environmental and agronomic variables. The dataset features include soil moisture, air temperature, relative humidity, solar radiation, and crop types, among other features, obtained using sensor networks installed on farmland. We trained and tested each model before comparing its performance using standard evaluation metrics, which include accuracy, precision, recall, F1 Score, and the Area Under the Curve. These findings indicate that GB and KNN models performed better than SVM and LR. For instance, GB and KNN achieved precisions of 95.6% and 92.4%, respectively, compared to SVM and LR, which achieved precisions of 86.2% and 72.8%, respectively. In both accuracy and generalization, the GB model performs overall best. This study contributes a fair investigation of the suitability of well-known ML models in irrigation forecasting for smart farming in the south-western region of Nigeria. This study makes use of a region-specific dataset that is gathered by sensor networks, involving 100,000 records in two farming seasons.