Accurate monitoring of micronutrients in tilled potato soils of eastern Canada: Application of an eXplainable inspired-adaptive boosting framework coupled with SelectKbest

•Experimental investigation on the nutrient metals in potato soils of eastern Canada.•An eXplainable MBA-AdaBoost coupled with SelectKbest to predict copper and zinc.•Optimization of huge physicochemical inputs using the SelectKbest feature selection.•Validating the primary model using MBA-CatBoost, classical AadaBoost, KRR, and MARS. Plant growth is significantly dependent upon the combination and concentration of mineral nutrients in the soil, where the adequate supply of these nutrients is a severe issue in fulfilling the fundamental cellular process requirements. This study collected 22 physicochemical properties of tilled potato soils from eight stations in two Atlantic Canadian provinces (Prince Edward Island and New Brunswick). Along with the experimental investigation, an explainable dual pre-processing inspired-intelligent paradigm comprised of the SelectKbest feature selection (FS), modified bat algorithm (MBA), adaptive boosting (AdaBoost), and Shapley Additive Explanations (SHAP) explainer was designed to monitor the micronutrients, including copper (Cu) and zinc (Zn). The significant data predictors were filtered using the SelectKbest FS to monitor Cu and Zn. The MBA was coupled with AdaBoost for tuning the hyperparameters to ensure accurate predictions. To validate the outcomes of the MBA-AdaBoost, four advanced machine learning approaches, including categorical boosting (CatBoost) coupled with MBA (MBA-CatBoost), classical AadaBoost, ridge kernel regression (KRR), and multivariate adaptive regression splines (MARS), were examined to compare the accuracies. The robustness of the developed framework and the performance of comparative models were examined through several statistical metrics. Results revealed that the MBA-CatBoost performed the best in predicting the Cu (R = 0.9425, U1 = 0.1020, and U22 = 0.2036) and Zn in the soil (R = 0.9454, U1 = 0.0942, and U2 = 0.1858) when compared with other models. Furthermore, the SHAP explainer interpreted the block-box main model during the training phase by introducing the CEC and Fe as significant predictors to monitor the soil Cu and Zn, respectively. These findings demonstrate a clear and robust relationship between the presented modeling approach and the accurate prediction of soil micronutrient concentrations. Accurate predictions of micronutrients using these advanced techniques can be used to tailor site-specific micromovement management in agriculture fields to reduce production costs, increa