Data-driven prediction of heavy metal bioconcentration factors in sunflower: A multifactor modeling study

Heavy metal (HM) contamination of soil threatens food security and ecosystem health. Sunflower phytoextraction is an environment-friendly option for managing contaminated soil. However, formulating phytoremediation strategies requires a reliable prediction of the bioconcentration factor (BCF) of HMs in sunflower, which remains difficult because of the nonlinear, multifactor controls on HM uptake, and sparse and incomplete datasets. In this study, we developed a machine learning (ML) framework to predict the BCF of HMs in sunflower. Multiple ML models combined with various data preprocessing techniques were systematically evaluated to identify an optimal modeling approach. The optimal categorical boosting model achieved a mean coefficient of determination of 0.60 on the test set, demonstrating satisfactory predictive performance and robust generalization to unseen data. Shapley additive explanation analysis identified soil HM concentration, sunflower components, and electronegativity as dominant predictors. Partial dependence plots further revealed a decreasing trend in BCF with increasing soil HM concentrations. Moreover, structural equation modeling suggested that variation in BCF was primarily associated with direct effects of HM and soil properties, while soil properties also played an important mediating role. These findings provide a transferable framework for BCF prediction and practical guidance for optimizing sunflower-based phytoremediation strategies across diverse HMs and soil conditions.