Machine Learning-Driven Prediction of Microplastic Aging Processes and Environmental Risk Assessment Across Multi-Media Systems.

Machine learning (ML) holds promise for reconstructing microplastic (MP) aging and assessing risks, but current studies rely on small-scale, accelerated laboratory datasets and single environmental medium models that miss cross-media transport and environmental interactions in real-world MP lifecycles. To realize its potential for reconstructing spatiotemporal aging trajectories and toxicological assessment of MPs, this perspective provides a paradigm shift in ML application from fragmented data-fitting to a holistic, privacy-preserving, physics-aware strategy. A novel probabilistic framework reconstructs the environmental history of field-sampled MPs through mechanistic fingerprinting, using Bayesian inference to reconcile multi-evidence signals and improve trajectory models for source attribution and risk assessment. Furthermore, we propose the TRACE framework (TRansport, Aging, Corona, Ecotoxicity), which moves beyond the isolated modeling of aging processes and toxicity endpoints. By integrating physics-informed models with causal discovery, TRACE captures the reciprocal feedback loops between physicochemical evolution and eco-corona formation, thereby mechanistically linking surface transformations to biological risks. To support this data-intensive architecture, we advocate for federated learning (FL) to dismantle privacy barriers. This approach facilitates secure, multi-institutional collaborative modeling without raw data exchange, harmonizing heterogeneous datasets. Ultimately, this cohesive strategy bridges laboratory-field disparities, moving toward predictive, evidence-based, and targeted mitigation efforts in global plastic pollution governance.