The paper proposes a novel hybrid prediction model for water pollutants based on the hybrid osprey optimization algorithm - least squares support vector machine (OOA-LSSVM). Unlike conventional LSSVM approaches, the key innovation lies in employing the OOA to automatically and effectively optimize the two critical parameters of LSSVM—the regularization parameter (C) and the kernel parameter (σ)—thereby enhancing the model’s generalization capability and prediction accuracy. The main factors influencing pollutant concentrations in lakes are identified through the Pearson correlation coefficient, with the most significant ones selected as input variables for the proposed model. The performance of the proposed OOA-LSSVM model is evaluated using statistical metrics (R², mean absolute error, mean absolute percentage error, root mean square error) and compared with the standard LSSVM. The results demonstrate that the OOA-LSSVM model achieves superior prediction accuracy and robustness, offering a reliable and efficient computational tool for forecasting water pollutants and supporting lake pollution control strategies.