DEVELOPMENT AND ANALYSIS OF IoT ENABLED MACHINE LEARNING MODELS FOR COST-EFFECTIVE SOIL MOISTURE ESTIMATION

Abstract

Soil moisture plays a pivotal role in agricultural productivity, water resource management, and climate regulation. Traditional methods for soil moisture estimation often fall short in addressing the spatial and temporal variability of soil conditions, necessitating modern, intelligent alternatives. This thesis explores the development and evaluation of cost effective, IoT-enabled machine learning models for accurate soil moisture prediction. First, an in-depth overview of recent developments regarding Long Short-Term Memory networks, encoder-decoder architectures, and multimodal systems integrating satellite imagery and meteorological readings is given to put the present technological context into perspective. From such readings, an applied system consisting of low-cost Internet of Things sensors that record multi-sensor data in terms of temperature, humidity, and rain is constructed. The data collected are subjected to preprocessing methods such as normalization and imputation before being presented to ensemble learning-based modeling to improve prediction performance. The proposed system is validated using performance metrics such as RMSE, MAE, and R², demonstrating superior accuracy and real-time applicability in field conditions. This work not only bridges the gap between theory and practice but also offers scalable solutions for precision irrigation and drought mitigation in the context of sustainable agriculture.