Efficient water management is a cornerstone of sustainable precision agriculture, yet traditional irrigation often suffers from over-irrigation due to static scheduling. This paper presents a robust IoT-enabled smart irrigation framework that leverages the ESP32 microcontroller and a suite of environmental sensors (soil moisture, DHT22, rain, and water flow) integrated with machine learning for dynamic decision-making. Unlike threshold-based systems, our approach utilizes a Random Forest classifier to predict irrigation needs based on multivariate environmental inputs, achieving a prediction accuracy of 94.2%. Real-time data is synchronized with the Thing Speak cloud platform, enabling remote monitoring and data-driven insights. Experimental results demonstrate a 35% reduction in water consumption compared to conventional methods while maintaining optimal soil moisture levels for crop growth.