Abstract
Increasing global water scarcity has intensified the need for sustainable and decentralized wastewater treatment solutions. Phytoremediation using plants such as Helianthus annuus offers an environmentally friendly option for greywater reuse, yet its effectiveness is often constrained by plant stress under high surfactant loads. In this study, we introduce an integrated approach that combines microbial-assisted phytoremediation with machine learning (ML) to both enhance and predict remediation performance. Plants were supplemented with a rhizospheric consortium of Bacillus subtilis, Pseudomonas fluorescens, and Trichoderma harzianum. Relative to the greywater-only treatment, the three-member treatment was associated with higher indole-3-acetic acid concentrations and higher superoxide dismutase and catalase activities, indicating an altered antioxidant-enzyme response. These measurements do not directly establish lower reactive oxygen species accumulation or oxidative damage. To move beyond purely empirical evaluation, we developed a Bio-AI predictive model based on a hybrid Random Forest algorithm that integrates biochemical indicators with environmental variables and semantic representations of qualitative descriptors. The computational Bio-AI dataset comprised 500 observations (N = 500). Within the evaluated dataset and specified validation procedure, the SBERT-augmented Random Forest model achieved an R2 of 0.96, compared with approximately 0.65 for the linear-regression baseline. A separately defined classification task yielded a ROC-AUC of 0.98 for the Bio-AI model. These results indicate improved predictive performance within the evaluated dataset; however, independent cross-site and cross-species validation is required before operational deployment. Preliminary cross-species evaluation further suggested transferability of the physiological features, although this finding should be interpreted as exploratory rather than definitive evidence of generalizability. By linking physiological responses with advanced computational modeling, this work presents a scalable and practical framework for optimizing phytoremediation systems in urban water management.