Integrating machine learning and reliability analysis: A novel approach to predicting heavy metal removal efficiency using biochar

Abstract

Soil contamination with heavy metals (HMs) presents critical environmental and public health risks due to their long-term persistence and tendency to bioaccumulate. Biochar has gained recognition as an effective amendment for HM immobilization, owing to its cost-effectiveness, environmental sustainability, and multifunctional properties. Nevertheless, consistent removal efficiency remains challenging to achieve due to the inherent variability of biochar and its interactions with complex environmental factors. This research introduces an advanced machine learning (ML) framework, utilizing deep forest (DF) algorithms, to predict and optimize the efficiency HM removal through biochar applications. The framework addresses key challenges by employing data imputation to manage missing information, data augmentation to overcome limitations of small datasets, and reliability analysis to assess predictive uncertainties, thereby improving the model’s reliability and generalization capability. The findings reveal that the DF model surpasses conventional ML approaches, achieving a testing dataset coefficient of determination (R²) of 0.88. Additionally, probabilistic reliability analysis offers valuable insights into the likelihood of reaching various levels of remediation efficiency (RE). For lower RE thresholds, such as 20–30 %, the model predicts a high probability (over 80 %) of substantial HM removal, confirming biochar’s effectiveness in mitigating contamination. However, as the target RE thresholds rise to moderate levels (50–70 %), the probability drops significantly (to below 5 %), highlighting the increasing difficulty of achieving higher remediation efficiencies. Furthermore, this study has developed an accessible and intuitive web-based application, enabling engineers to input relevant parameters and receive immediate predictive outputs, thus facilitating the practical application of advanced ML models in real-world scenarios.