The escalating global water crisis necessitates a transition from traditional irrigation methods to data-driven precision agriculture. This study proposes an integrated framework combining Internet of Things (IoT) sensor networks with Machine Learning (ML) algorithms to optimize water usage in semi-arid agricultural zones. We deployed a network of soil moisture, temperature, and humidity sensors across a 10-acre test plot at Green Valley State College. Data was processed using a Random Forest Regressor to predict irrigation needs 24 hours in advance. Our results indicate a 22% reduction in water consumption and a 12% improvement in crop yield compared to traditional timer-based systems. This research demonstrates that affordable, localized IoT solutions can provide a scalable pathway for small-scale farmers to adopt sustainable practices