Due to the rapid global urbanization and industrial development, a substantial amount of untreated wastewater containing heavy metals is discharged into rivers and oceans. This has resulted in severe heavy metal pollution in the environment (Lan et al., 2020; Tan et al., 2021; Yan et al., 2019). Heavy metals such as lead, zinc, chromium, cadmium, and copper, can have harmful effects on the animals, plants, and humans within ecosystems due to their non-degradability and toxicity (Bashir et al., 2020; Kara et al., 2017). The escalating issue of heavy metal pollution has gained international attention, prompting urgent research in the field of heavy metal pollution prevention and control as a pressing global concern.
Currently, there are three main classes of methods for remediating heavy metal pollution in water: (I) Biological methods, including phytoremediation and microbial remediation; (II) Chemical methods, including precipitation, ion-exchange, and electrochemical remediation; (III) Physical methods, including dilution, adsorption, and membrane filtration (Chai et al., 2021; Gu et al., 2019; Tan et al., 2021; Tohdee et al., 2018). Adsorption is considered as an effective method for the remediation of heavy metal pollution owing to its high efficiency, cost-effectiveness, and operational convenience (Chai et al., 2021; Uddin, 2017). Adsorbents such as activated carbon and metal-organic frameworks (MOFs) exhibit advantages including high specific surface area, porous structures, and excellent adsorption performance, rendering them suitable for the adsorption treatment of heavy metals (Li et al., 2021; Peng et al., 2018; Qiu et al., 2021; Xu et al., 2021). However, their relatively high cost and complex preparation processes limit their practical engineering applications. In the pursuit of economically viable, readily available, and environmentally friendly adsorbents, researchers have developed and investigated natural adsorbents for their application in the remediation of heavy metal pollution (Dhar et al., 2023; Gu et al., 2019). Bentonite is a type of natural layered clay that is primarily composed of montmorillonite. It has a large specific surface area and significant amounts of permanent negative charges in its interlayer spaces, which makes it highly effective in adsorbing heavy cations through ion-exchange (Tohdee et al., 2018). Furthermore, with abundant reserves, low cost, and natural harmlessness, bentonite has great potential in practical applications. Many scholars have proved the effectiveness of bentonite in adsorbing heavy metals, and the adsorption behaviors and mechanisms have been thoroughly investigated (Chen et al., 2012; Hamidpour et al., 2010; Kaya and Ören, 2005a). However, there is currently a lack of a universal predictive model for the heavy metal adsorption capacity of bentonite. This deficiency may lead to inferior efficiency, increased costs, and material wastage when employing bentonite for the treatment of heavy metal-containing wastewater in practical applications. Therefore, in order to enhance the application of bentonite in adsorbing heavy metals, it is imperative to propose a predictive model or method for the heavy metal adsorption capacity of bentonite. While traditional orthogonal experimental design and response surface methodology can produce a predictive model, it is important to note that such models are only reliable for specific experimental conditions and exhibit poor generalization performance (Huo et al., 2022). Given the diversity in the bentonite properties and adsorption conditions in practical applications, traditional methods are unable to effectively learn the nonlinear relationships among extensive datasets, making it challenging to predict the adsorption the heavy metal adsorption capacity of bentonite accurately and reliably.
Machine learning (ML) is a method involving computer science and artificial intelligence that can learn from empirical data to capture intricate nonlinear relationships and construct highly accurate regression predictive models (Zhang et al., 2023; Zhao et al., 2022). The knowledge acquired by ML models during training can typically generalize to unseen data, thereby enhancing the model’s performance in practical applications. In recent years, ML has been widely employed in the field of environmental science, including predicting the heavy metal adsorption capacity of biochar (Zhu et al., 2019), urban carbon dioxide emissions (Qin and Gong, 2022), the migration of radionuclide solutes in groundwater (Meray et al., 2022), and the carbon dioxide capture capacity of MOFs (Orhan et al., 2023). ML models can be used to predict experimental outcomes, guide the optimization of experimental designs, and reduce redundant experiments, thereby enhancing experimental efficiency and resource utilization (Huo et al., 2022). In different practical tasks related to the adsorption of heavy metals, many features that influence the heavy metal adsorption capacity of bentonite include the variation in bentonite properties, diversity in adsorption conditions, and differences in target heavy metals. Traditional experimental methods require significant resources and time to construct intricate models and often exhibit poor generalization capabilities. But ML is highly suitable for learning from these multidimensional features (Palansooriya et al., 2022), enabling the establishment of accurate regression predictive models to enhance the effective application of bentonite in adsorbing heavy metals. Given the current absence of accurate predictive models for the heavy metal adsorption capacity of bentonite, it is important to employ ML to develop a predictive model with high accuracy and generalization capacity.
The aim of this study is to investigate the heavy metal adsorption of bentonite by using ML algorithms and to develop an accurate predictive model for the heavy metal adsorption capacity of bentonite. The dataset was compiled by extracting samples from publicly available literature on the heavy metal adsorption of bentonite. Six ML algorithms were used to train on the dataset with the aim of developing an optimal predictive model. Experimental validation was conducted to evaluate the model’s generalization capacity, ensuring its accuracy and robustness across diverse datasets and practical scenarios. The programming language employed in this study was Python (version 3.11.5). The findings of this study can be utilized for the rapid prediction of heavy metal adsorption capacity of bentonite, holding significant implications for understanding the importance of various features, adjusting experimental designs, and contributing to environmental remediation.