Precision irrigation is an essential component of optimizing water use for sustainable agriculture. This paper describes an IoT-based irrigation system that incorporates several environmental sensors to track important parameters like soil moisture, temperature, pH, air humidity, wind speed, and evapotranspiration rates. The sensor data are processed by a PID controller, which controls the operation of water pumps automatically in accordance with real-time irrigation requirements. The sensor readings are saved in a cloud platform, where sophisticated neural networks such as Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Temporal Fusion Transformer (TFT), and Multi-Layer Perceptron (MLP) are trained to forecast the best water needs. The models are tested using precision, recall, F1-score, accuracy, root mean square error (RMSE), mean absolute error (MAE), and computational efficiency. Findings show that the TFT model is the most accurate in prediction but consumes a lot of computational power, and thus it is best suited for cloud-based systems. The GRU and LSTM models offer a trade-off between efficiency and accuracy, and thus they are best for real-time applications in smart irrigation systems. This study illustrates the capability of deep learning and IoT technologies to improve irrigation methods, utilizing water efficiently while keeping crops in their best growth conditions. Through the use of real-time data and predictive analytics, this method supports sustainable agriculture, minimizing water loss and optimizing resource allocation.