<![CDATA[Soil moisture is a critical parameter in agricultural and hydrological systems that requires accurate monitoring to support smart irrigation management. This study aims to develop a soil moisture forecasting model based on stacking ensemble machine learning using time series data from an Arduino-based Data Acquisition (DAQ) system. The dataset includes environmental variables such as atmospheric temperature, soil temperature, humidity, soil_moisture, and dew point with hourly temporal resolution over the 2017–2018 period. The research stages include data preprocessing (missing value handling, interpolation, outlier handling using the IQR method, and resampling), feature engineering (temporal feature extraction, lag features, and rolling windows), and modeling using six tree-based methods: Random Forest, Gradient Boosting, Extra Trees, LightGBM, XGBoost, and CatBoost. The three best-performing models (CatBoost, LightGBM, and Gradient Boosting) were combined using Ridge Regression as a meta-learner. Evaluation was conducted through time-series cross-validation with MSE, RMSE, MAE, R², and MAPE metrics. Test results show the Ridge Ensemble achieved MSE of 52.21, RMSE of 7.23, MAE of 5.16, R² of 0.9652, and MAPE of 9.07%. The stacking ensemble approach outperformed individual models and shows strong potential for real-time deployment in IoT systems to support precision agriculture.]]>
Copyrights © 2026
