Bayesian optimization with Optuna for enhanced soil nutrient prediction: a comparative study with genetic algorithm and particle swarm optimization
Bamidele Abayomi Dada, Nnamdi Nwulu, Seun O. Olukanmi
Abstract
Optimizing soil nutrient prediction models is important for achieving maximum agricultural output and sustainability while also ensuring effective resource management and environmental protection, as demonstrated by a case study in Johannesburg, South Africa. We implemented machine learning (ML), optimization, geographic information systems, and remote sensing. This research investigates the effectiveness of ML algorithms, including random forest (RF), Adaboost (ADB), gradient boosting (GB), and XGBoost (XGB), when used with high-resolution earth observation data. In addition, it examines 2,000 random surface soil samples, ranging from 0 to 20 cm, that were optimized using genetic algorithms (GA), particle swarm optimization (PSO), and Optuna. We train them with 70% of the data. The investigation confirms that Optuna-optimized models are at least 13% more precise than GA and PSO models. The concordance correlation coefficient (CCC), R-squared (R²), and mean absolute percentage error (MAPE) increased, while the root mean squared error (RMSE) and mean absolute error (MAE) decreased. Optuna's tree-structured Parzen estimator (TPE) and pruning algorithms are employed to generate more precise estimates of soil nutrients. The majority of models are reduced, computation is expedited, and hyperparameters are enhanced. In the context of precision agriculture, these developments are directly applicable because they enable data-driven fertiliser management, reduce waste, and increase yields. Improved nutrient prediction is also advantageous from an environmental perspective, as it reduces the need for superfluous fertilizer applications and prevents discharge caused by excess fertilizers. Further research will be conducted on reinforcement learning for adaptive searching, multi-objective optimization, and the facilitation of hyperparameter tuning to develop more precise models for predicting soil nutrients.