A Multi-Task Stacked Ensemble and IoT-Enabled Decision Support System for Precision Fertigation in Smallholder Agriculture
Year: 2026
Type: article
Abstract: Nigerian agriculture's fixed-schedule fertigation causes low efficiency and nutrient leaching.A stacked-ensemble model is developed for precision fertigation that jointly predicts fertigation need, rate (kg/ha) and timing (Early/Optimal/Late).To train and evaluate the ensemble, a unified dataset was integrated, comprising 12,840 records and 42 variables from a Nigerian soil-weather-yield dataset, a locally sourced Nigerian IoT sensor series and historical weather/NDVI feeds.An LSTM soil-dynamics model, an XGBoost rate regressor and Random Forest need/timing classifiers are fused through an XGBoost meta-learner trained on out-of-fold predictions.On held-out partitions, the ensemble reduced rate MAE from 0.55 to 0.49 kg/ha (-10.9%) and RMSE from 0.68 to 0.61 kg/ha (-10.3%;R² 0.88→0.92),raised need F1 from 0.83 to 0.86 (accuracy 0.87→0.89;AUC 0.89→0.93)and timing macro-F1 from 0.84 to 0.86, with well-calibrated probabilities (Brier 0.082).The trained ensemble was deployed through a RESTful API and responsive dashboard; under concurrent load, the system recorded 0% request errors with 1.88s median API latency, demonstrating practical deployability for Nigerian smallholder agriculture.
Authors: Awojide S., Ikpotokin F.O., Sadiq F.I.
Language: English
Topic: Smart Agriculture and AI
Subfield: Plant Science
Domain: Life Sciences
Open Access status: diamond
OpenAlex ID: W7213663876