This study presents an IoT-enabled smart irrigation management system utilizing subsurface soil moisture sensors and a recurrent neural network–long short-term memory (RNN-LSTM) model to predict soil moisture in real-time for precision agriculture. The proposed system was deployed in Malaysia for six months, achieving a root mean square error (RMSE) of 1.222, a mean absolute error (MAE) of 0.6374, and a coefficient of determination (R²) of 0.6723, explaining approximately 67% of the variance in the observed data. Additionally, 95.49% of predictions fell within ±5% of actual measured values, a tolerance-based metric distinct from classification accuracy. Outlier analysis revealed that the largest residuals occurred during heavy rainfall events, and adopting a robust Huber loss function improved R² to 0.70. The results indicate that the system can effectively support irrigation scheduling, although future work should extend seasonal coverage and address spatial variability in larger fields.
Maniam et al. (2026) studied this question.