Efficient irrigation management is crucial for sustainable agriculture, especially in arid regions where water scarcity presents significant challenges. This study evaluated the performance of eight machine learning (ML) models-Linear Regression (LR), Random Forest (RF), XGBoost, Support Vector Regression (SVR), K-Nearest Neighbors (KNN), ARIMA, Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU)-to predict soil moisture (SM) for sugar beet cultivation in southern Algeria. Using climatic data and soil moisture readings from a patented smart irrigation system, the models were assessed based on Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and the correlation coefficient (R). The results showed that LSTM and GRU models achieved the highest accuracy. Specifically, the GRU model yielded an R value of 0.82, an RMSE of 0.987, and a MAE of 0.747, while the LSTM model yielded an R value of 0.77, an RMSE of 0.998, and a MAE of 0.826. Both models effectively captured the temporal and nonlinear dynamics of the data. In contrast, traditional models like LR and K-Nearest Neighbors (KNN) performed poorly, with R values of 0.30 and 0.13, respectively. Integrating ML predictions with the Penman-Monteith equation further enhanced irrigation scheduling, resulting in an estimated 15-25% reduction in water usage, and a 5-10% increase in crop yield potential, while maintaining optimal crop hydration.These findings underscore the transformative potential of ML techniques in optimizing irrigation systems, advancing sustainable agricultural practices in water-scarce regions.