ML Model Training

Train regressors to predict continuous WQI. Metrics: MAE, MSE, RMSE, R².

Optimization uses bounded cross-validation; Bayesian optimization runs 20 trials and may take longer than normal training.