Explainable machine learning for predicting diarrhetic shellfish poisoning events in the Adriatic Sea using long-term monitoring data

In this study, explainable machine learning techniques are applied to predict the toxicity of mussels in the Gulf of Trieste (Adriatic Sea) caused by harmful algal blooms. By analysing a newly created 28-year dataset containing records of toxic phytoplankton in mussel farming areas and diarrhetic shellfish toxins in mussels (Mytilus galloprovincialis), we train and evaluate the performance of machine learning (ML) models to accurately predict diarrhetic shellfish poisoning (DSP) events. Based on the F1 score, the random forest model provided the best prediction of toxicity results at which the harvesting of mussels is stopped according to EU regulations. Explainability methods such as permutation importance and Shapley Additive Explanations (SHAP) identified key species (Dinophysis fortii and D. caudata) and environmental factors (salinity, river discharge and precipitation) as the best predictors of DSP toxins above regulatory limits. These findings are important for improving early warning systems, which until now were based solely on empirically defined alert abundances of DSP species. They provide experts, aquaculture practitioners, and authorities with additional information to make informed risk management decisions.