Comparative analysis of machine learning and novel deep learning models for IoT enabled smart irrigation water management system

Efficient irrigation techniques are critical for sustainable agriculture, given the increasing demand for limited water resources and the need for higher productivity. Traditional irrigation methods often lack precision, resulting in water waste and reduced crop yields. This study presents a measurement-based irrigation system that leverages the Internet of Things (IoT) alongside machine learning and deep learning methodologies. IoT sensor data from agricultural fields provides real-time insights into soil and environmental conditions, facilitating precise irrigation decisions. The proposed framework employs Random Forest ensemble methods, including stacking and bagging, in conjunction with a BiLSTM model to predict pump status and determine optimal water requirements. Experimental results indicate that stacking models substantially outperform single-model approaches. This methodology integrates machine learning into measurement science, encompassing sensor calibration, performance benchmarking, and agronomic validation, rather than restricting it to post hoc analysis. The system delivers accurate forecasting and control, enabling real-time application in smart agriculture. Its scalability and flexibility make it a promising solution for sustainable water management and improved crop yields.