Prediction of fecal coliform concentration in coastal surface water using machine learning algorithms and water quality parameters.

Fecal coliform (FC) is a key indicator of surface water contamination from animal and human waste, and elevated FC levels are associated with waterborne diseases, which remain major public health concerns. However, existing studies are often limited by sparse sampling, time-consuming laboratory analyses, and inadequate representation of complex nonlinear relationships among water quality parameters (WQPs) influencing FC dynamics. Therefore, this study aims to apply and compare machine learning models for predicting FC concentrations using physicochemical WQPs, including pH, salinity (WS), water temperature (WT), electrical conductivity (EC), total dissolved solids (TDS), dissolved oxygen (DO), turbidity (Turb), biochemical oxygen demand (BOD), and water hardness (WH), while evaluating their predictive contributions and interactions in coastal Bangladesh. FC concentrations ranged from 188 to 1620 CFU/100 mL, with a mean of 765.57 CFU/100 mL. Pearson correlation analysis ranked the predictors as Turb > BOD > DO > pH > WS > WH > WT > TDS > EC. Among the models, the multilayer perceptron (MLP) outperformed random forest regression (RFR), multiple linear regression (MLR), extreme gradient boosting (XGB), and support vector regression (SVR). During testing, MLP achieved the highest R2 (0.822), with the lowest RMSE (109.48 CFU/100 mL) and MAE (99.95 CFU/100 mL). Notably, the optimal input combination (IC-5), comprising only five WQPs (Turb, BOD, DO, pH, and WS), explained 80.8% of the variability in FC concentrations and achieved 98.30% of the R2 obtained using all nine WQPs. These findings may provide a useful basis for microbial water-quality assessment in comparable coastal environments.