Answers are grounded in this paper and cite their sources.
Cyanobacterial harmful algal blooms (CyanoHABs) pose significant risks to freshwater ecosystems, necessitating accurate and interpretable forecasting approaches. This study develops seven machine learning (ML) models, namely Extreme Gradient Boosting (XGB), Gradient Boosting Machine (GBM), Artificial Neural Network (ANN), Random Forest (RF), Ridge Regression (Ridge reg.), and Support Vector Regression with radial bias function and linear kernels (SVR-rbf, SVR-linear), to predict phycocyanin concentration in the Lower Charles River using high-frequency monitoring data. Predictors were selected using the Boruta algorithm, and datasets were chronologically split into training (80%), validation (10%), and testing (10%) sets. XGB achieved the highest predictive accuracy (average r = 0.965, RMSE = 0.197, MAE = 0.147, and MAPE = 9.373%), while SVR-linear showed the lowest. Boruta identified water temperature, specific conductivity, and chlorophyll-a as key predictors. SHAP based explainable artificial intelligence (XAI) further quantified feature contributions, identifying specific conductivity, chlorophyll-a, and turbidity as dominant drivers governing bloom dynamics and revealing nonlinear threshold behavior. The proposed framework integrates feature selection with explainable AI, demonstrating the superiority of boosting-based model while providing both high predictive accuracy and mechanistic interpretability. This approach offers advanced tools for real-time HAB prediction and enables data-driven early bloom warning system for water quality management.
No takes yet. Share an insight, caveat, or question.
Mahat et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: