UPMO: An Uncertainty-Aware Predictive Multi-Objective Optimization Framework for Smart Agriculture

Abstract

Smart agriculture and agro-industrial systems increasingly rely on sensor-driven data to optimize resource utilization while maintaining high productivity. However, these systems involve inherently conflicting ob j ectives, par ticularly minimizing energy consumption while maximizing yield efficiency. This paper presents UPMO, an uncertainty-aware predictive multi-objective optimization framework that integrates Gaussian Process Regression (GPR) with the Non-dominated Sorting Genetic Algorithm II (NSGA-II). The framework is evaluated on a real-world environmental sensor telemetry dataset containing over 130,000 observations. Derived target variables representing energy consumption and yield efficiency a rein troduced to en ableop timizatn. T pipeline combines K-nearest neighbors (KNN) imputation, temporal feature engineering, and Random Forest-based feature selection to identify key control variables. Gaussian Process models with RBF and white-noise kernels are used as surrogate models to capture nonlinear dynamics while providing predictive uncertainty. These models are embedded within NSGA-II to generate Pareto-optimal solutions under an uncertainty-aware objective formulation. Experimental results demonstrate strong predictive performance, achieving MAE values of 0.76 for energy and 0.63 for yield, with $R^{2}$ up to 0.67. The Pareto frontier reveals a clear trade-off between energy (36.0-39.7) and yield (92.7-95.0%), enabling identification of optimal o perating r egimes. The framework achieves a hypervolume of 0.81 and spacing of 0.024, indicating strong convergence and solution diversity. From an operational perspective, the proposed approach identifies e n e rgy-efficient co nfi gur ations red uci ng consumptio by up to 8-12% while maintaining yield above 93%. Overall, UPMO provides a scalable, interpretable, and robust decisionsupport framework bridging predictive modeling and real-world optimization in smart agriculture systems.