Explainable machine learning for predicting microplastic burden in commercially important freshwater fish and assessing dietary exposure

Microplastic (MP) contamination in commercially important freshwater fish represents an emerging concern for aquatic ecosystem integrity and human food safety. This study presents an interpretable machine learning framework for predicting and assessing MP accumulation in European sea bass (Dicentrarchus labrax), Nile tilapia (Oreochromis niloticus), bighead carp (Hypophthalmichthys nobilis) and crucian carp (Carassius auratus) collected from four retail markets in Yichang City, Hubei Province, China. A total of 480 fish specimens generated 980 analytical records following gastrointestinal tract (GIT) extraction by alkaline digestion using 10% potassium hydroxide (KOH). Recovered MPs were characterised using attenuated total reflectance Fourier-transform infrared (ATR-FTIR) spectroscopy and scanning electron microscopy coupled with energy-dispersive X-ray spectroscopy (SEM-EDS). Six supervised regression algorithms, namely XGBoost, LightGBM, CatBoost, Random Forest, Support Vector Regression (SVR) and Artificial Neural Network (ANN), were evaluated using an 80:20 stratified training–testing split (784/196 records). XGBoost achieved the highest predictive performance (R2 = 0.96, RMSE = 0.48 MPs kg−1) and was selected as the optimal model. SHapley Additive exPlanations (SHAP) identified fish weight, total length, market of origin and species as the principal predictors of MP burden. Polypropylene (PP), polyethylene (PE) and polystyrene (PS) predominated, with fibres representing the dominant morphology. Estimated dietary exposure reached approximately 4,650 particles year−1 for adult males, while 32% of fish exceeded 0.04 MPs g−1 wet weight. Because MPs were measured exclusively in GIT tissue, this estimate represents a conservative upper-bound approximation of dietary exposure.