Polymer‑specific hazard profiling and risk indexing of microplastics in coastal sediments of St. Martin’s Island: A multivariate and machine learning approach

• This study presents the first polymer-resolved evaluation of microplastic contamination in surface sediments from a small island ecosystem in the Bay of Bengal. • Microplastic assemblages were dominated by fibres and fragments, with polyethylene and polypropylene most abundant, reflecting persistent inputs from fishing gear and ropes. • We propose a novel Microplastic Pollution Risk Index (MPRI) that integrates polymer hazard scores with persistence, morphology, and color, providing a comprehensive hazard profile. • Tourism-exposed beaches exhibited higher diversity and shares of polystyrene, PET, and PVC, linking packaging and consumer waste to localized contamination hotspots. • Multivariate analyses (Pearson correlation, PCA, RDA, HCA) and machine learning models (Random Forest, SVM, KNN) clearly separated tourism and fishing sites, underscoring distinct contamination pathways. Microplastic (MP) contamination poses an emerging ecological threat in small-island environments, yet polymer-specific risk assessments remain limited for the Bay of Bengal region. This study provides the first integrated, polymer-resolved evaluation of microplastics in coastal sediments of St. Martin’s Island, Bangladesh. A total of 298 suspected MPs were isolated through stereomicroscopy, of which 250 particles were confirmed through ATR-FTIR (≥89% spectral match). Fibres and fragments dominated the assemblage, with high abundances observed in tourism-intensive beaches (S1–S4, S12) and fishing-dominated zones (S8–S11). Polymer profiles were characterized by the predominance of PE, PP, PET, and PVC. Four complementary ecological risk metrics, Pollution Load Index (PLI), Polymer Hazard Index (PHI), Sediment Polymer Hazard Index (SPHI), and the Microplastic Pollution Risk Index (MPRI), identified localized hotspots of elevated risk, particularly at tourism and active fishing sites. Multivariate analyses (PCA, HCA) revealed clear clustering patterns associated with site-use categories, while machine-learning classifiers (Random Forest, SVM, KNN) accurately distinguished tourism, fishing, and low-use zones based on MP morphology, color, polymer type, and abundance. Collectively, these findings demonstrate that anthropogenic pressure strongly shapes microplastic composition and hazard profiles on St. Martin’s Island. The integrated risk-index and ML framework presented here provides a robust, reproducible approach for coastal microplastic monitoring in data-limited regions and can support targeted management and mitigation strategies in vulnerable island ecosystems.