Monitoring microbial water quality at recreational beaches is essential to safeguard public health, with fecal coliforms serving as key indicators of contamination. This study applies machine learning (ML) techniques to predict fecal coliform concentrations at Montevideo’s urban beaches, aiming to support proactive and data-driven coastal water quality management. Using an extensive monitoring dataset, we developed and calibrated five ML models to predict continuous fecal coliform levels, improving upon traditional threshold-based methods. Among these, Random Forest (RF) and Histogram-based Gradient Boosting (HGB) models showed very good predictive performance, with RF yielding the most consistent estimates of microbial contamination and HGB showing comparable accuracy but higher predictive uncertainty. The models were optimized using cross-validation and Optuna, with mean squared error as the loss function. Feature importance analysis using SHAP values revealed that Enterococcus concentrations were the most influential predictor, followed by water temperature and salinity. Seasonal patterns in coliform levels were also identified, likely linked to fluctuations in water temperature. These findings provide actionable insights into the dynamics of microbial contamination and highlight the potential of ML models for early warning systems, adaptive monitoring, and improved risk communication. This integrative approach not only enhances predictive performance but also advances our understanding of the environmental processes influencing water quality in urban coastal systems.