Decoding the microplastic crisis: Machine learning and deep learning in transforming data into ecological insight

The global prevalence of microplastics (MPs) presents a pressing environmental challenge, yet assessing their ecological impact is constrained by the labor-intensive and subjective nature of conventional analytical protocols. Artificial intelligence, encompassing machine learning (ML) and deep learning (DL), has emerged as a powerful computational paradigm to automate polymer identification from complex Raman and FTIR spectra, while advanced segmentation architectures (e.g., U-Net, Mask R-CNN) enable high-throughput morphological profiling from microscopy images. Beyond characterization, ML models increasingly serve as prognostic engines to predict spatiotemporal transport, contaminant adsorption kinetics, and ecotoxicological risk endpoints. Despite these notable advancements, critical methodological bottlenecks persist, including conditional algorithmic performance trade-offs, pervasive dataset biases toward pristine polymers, matrix-induced false positives, data leakage from improper validation, and the opacity of "black box" models. While explainable AI (XAI) tools like SHAP and LIME provide valuable diagnostic insights into influential features, their post-hoc surrogate approximations and current lack of uncertainty quantification require critical interpretation. Moving forward, integrating multimodal data fusion, unsupervised discovery architectures, and lightweight Edge AI systems holds substantial potential to bridge the lab-to-field gap and establish an evidence-based foundation for real-time environmental monitoring and regulatory policy.