Machine Learning Steers a Biodegradable Film That Scrubs Textile Dye From Water

Textile dye pollution is one of the most stubborn problems in industrial wastewater treatment, and a new study from Ubon Ratchathani University in Thailand suggests that the answer may lie in pairing humble, biodegradable materials with one of machine learning’s most reliable workhorses. In research published in Environmental Science and Pollution Research, Chatchai Kunyawut and Chakkrit Umpuch describe a thin-film composite made from cross-linked gelatin and alginate reinforced with tetradecyltrimethylammonium bromide-modified montmorillonite clay, and show that an optimized gradient boosting model can predict and optimize its performance in removing Reactive Red 120, a widely used anionic azo dye, with remarkable precision.

The material itself is a clever piece of green chemistry. Gelatin, a protein derived from collagen, and alginate, a polysaccharide extracted from brown seaweed, are both abundant, biodegradable, and rich in functional groups that can interact with dissolved pollutants. By cross-linking the two polymers, the researchers created a stable film that resists dissolving in water, a common failure mode for purely biopolymer adsorbents. Embedded within this matrix are particles of montmorillonite, a naturally occurring swelling clay whose negatively charged layers normally attract cations rather than the anionic dye molecules that dominate reactive dye effluents.

That is where the surfactant comes in. Treating the clay with tetradecyltrimethylammonium bromide, a cationic surfactant known as TTAB, swaps the naturally occurring counterions between the clay layers for long, positively charged alkylammonium chains. This organophilic modification transforms the clay interior into an environment that welcomes anionic species such as RR120, whose sulfonate groups are drawn electrostatically to the quaternary ammonium sites. The result is a hybrid in which the clay supplies high-affinity binding sites while the surrounding biopolymer network holds everything together in a form that can be handled, recovered, and ultimately degraded.

The most consequential design decision, however, was architectural. The team had previously reported a bulk version of the same composite, but diffusion through thick granular adsorbents is slow: dye molecules must wind through tortuous internal pores before reaching binding sites buried deep within the material. By casting the composite as a thin film, the researchers dramatically shortened the diffusion pathways. Less time spent traveling means less internal mass-transfer resistance, and more of the material’s binding capacity becomes accessible within a practical contact time. It is a reminder that in adsorption science, geometry can matter as much as chemistry.

Batch experiments mapped out how the system behaves under varying conditions. Equilibrium adsorption capacity rose with contact time, initial dye concentration, and temperature, but fell as solution pH increased. That pH dependence is chemically coherent: under acidic conditions the adsorbent surface carries a more positive charge, strengthening the electrostatic pull on the anionic dye, while at higher pH, hydroxide ions compete with dye molecules for cationic sites and the surface charge shifts unfavorably. The positive temperature effect points to an endothermic adsorption process, in which thermal energy helps dye molecules overcome barriers to reaching and anchoring at active sites.

To dissect the mechanism, the researchers fitted their kinetic data to three classical models: pseudo-first-order, pseudo-second-order, and intraparticle diffusion. The analysis indicated that uptake was governed by a combination of surface adsorption and intraparticle diffusion, meaning that dye molecules both bind rapidly at external sites and gradually penetrate the film interior. At equilibrium, the Langmuir isotherm described the data best, implying monolayer adsorption onto a finite set of energetically equivalent sites, with a maximum adsorption capacity of 46.79 milligrams of dye per gram of composite. Regeneration tests using 0.5 molar sodium chloride showed moderate reusability, with chloride ions displacing bound dye and allowing the film to be cycled again.

The study’s second act is what elevates it beyond a conventional adsorption paper. Rather than relying solely on laborious one-factor-at-a-time experiments, the team trained several machine-learning algorithms on their experimental dataset and asked which could best predict adsorption capacity from the operating variables. Among the candidates evaluated, an optimized gradient boosting model emerged as the clear winner, achieving a coefficient of determination of 0.9606 on the test set and a root-mean-square error of just 1.4265 milligrams per gram. Gradient boosting works by building an ensemble of weak predictive models, typically shallow decision trees, each trained to correct the residual errors of its predecessors, and it has become a favorite for small-to-medium experimental datasets in environmental engineering because it captures nonlinear interactions without enormous data volumes.

With a trustworthy model in hand, the researchers used it to search the operating space for the conditions that would maximize dye uptake. The algorithm pointed to a contact time of 356 minutes, an initial dye concentration of 277 milligrams per liter, a pH of 2.8, and a temperature of 41 degrees Celsius. The crucial test came next: the team ran actual batch experiments at those predicted optima and measured an adsorption capacity of 45.23 plus or minus 0.05 milligrams per gram, a relative prediction error of only 0.91 percent. That level of agreement between simulation and experiment is the kind of result that turns a modeling exercise into a genuine engineering tool.

The broader significance is twofold. First, the work demonstrates that biodegradable, low-cost adsorbents need not sacrifice performance, particularly when smart design choices such as thin-film geometry and surfactant intercalation are deployed to overcome the classic weaknesses of biopolymer systems. Second, it shows that machine learning can compress the optimization workflow for water treatment processes: instead of exhaustively mapping every combination of time, concentration, pH, and temperature in the lab, researchers can train a model on a manageable dataset and let the algorithm navigate toward the optimum, reserving experiments for validation. Similar approaches are spreading rapidly across adsorption science, from biochar design for heavy metal capture to carbon materials for carbon dioxide separation.

Challenges remain before such films could see real-world deployment. The moderate reusability observed with salt regeneration suggests that long-term cycling performance would need improvement, and real textile effluents carry a far messier cocktail of salts, surfactants, and competing organics than a synthetic dye solution. Still, the study offers a compelling template: a biodegradable polymer-clay hybrid engineered for fast mass transfer, a classical mechanistic framework that explains why it works, and a machine-learning layer that tells operators exactly how to run it. As textile industries face tightening discharge standards and growing scrutiny of azo dye toxicity, strategies that combine sustainable materials with data-driven optimization are likely to move from the laboratory bench toward the treatment plant, and this thin-film composite with its gradient boosting guidepost is a vivid example of what that transition can look like.

Subject of Research: Machine learning optimization of Reactive Red 120 dye biosorption onto a biodegradable gelatin/alginate–montmorillonite thin-film composite

Article Title: Machine learning for optimizing Reactive Red 120 biosorption onto cross-linked gelatin/alginate-TTAB-montmorillonite composite

Article References: Machine learning for optimizing Reactive Red 120 biosorption onto cross-linked gelatin/alginate-TTAB-montmorillonite composite. (n.d.). https://doi.org/10.1007/s11356-026-38211-y

Image Credits: AI Generated

DOI: 10.1007/s11356-026-38211-y

Keywords: Reactive Red 120, biosorption, machine learning, gradient boosting, gelatin, alginate, montmorillonite, TTAB, thin-film composite, textile wastewater, adsorption, Langmuir isotherm