Predicting plant uptake of potential contaminants of emerging concerns using machine learning models (2018–2025): A global review

Abstract

This comprehensive review (2018–2025) synthesizes studies conducted on machine learning (ML) models to predict plant uptake of contaminants of emerging concerns (CECs). CECs are a global concern driven by anthropogenic activities, and their plant uptake is critical for environmental remediation and food safety. ML offers powerful predictive opportunities, fostering ecosystem preservation, human health and sustainable development. The review analyzes diverse ML algorithms, including ensemble models (random forest, gradient boosted regression trees, eXtreme gradient boosting) and deep learning (deep neural networks, recurrent neural networks, long short-term memory) alongside key predictors like soil pH, organic matter, and plant traits. It identifies dominant predictors and modelling approaches while highlighting significant research gaps: limited data, inconsistent feature reporting, and underexplored uncertainty-sensitivity coupling. A notable geographic imbalance exists, with China dominating research (82.1 % of 28 studies), followed by the USA (14.3 %) and Italy (3.6 %). Africa, particularly South Africa, is significantly underrepresented despite prevalent CEC pollution in African water bodies, including those used for irrigation, and agricultural vulnerabilities. This reveals a critical gap in global efforts to address crop contaminant risk, necessitating model retraining or transfer learning for African contexts given existing models' China-centric training. The findings underscore the urgent need for standardized databases, global sensitivity and uncertainty analysis and expanded geographic representation to enhance model generalizability and interpretability. This synthesis bridges environmental chemistry, plant physiology and artificial intelligence in predictive contaminant modelling, providing a foundation for future, globally relevant research.