Machine Learning-Based Prediction of 2-MIB Generation Importance Assessment of Water Quality Factors

Odor contamination in drinking water sources poses significant challenges due to compounds such as 2-methylisoborneol (2-MIB), which has an extremely low odor threshold. This study develops an automated machine learning (AutoML) approach to predict 2-MIB concentrations in a source reservoir and identify key influencing water quality factors. Using 470 samples covering 13 physicochemical and biological indicators from a reservoir in North China, the TPOT AutoML framework generated an optimal modeling pipeline that combined data preprocessing, feature engineering, and regression. The best model achieved high predictive performance ( R 2 = 0.946 in training; 0.734 in testing), while interpretability analysis and correlation analysis indicate that increased algal density was associated with synchronous increases in total organic carbon (TOC) and 2-methylisoborneol (2-MIB). Elevated TOC concentrations were statistically associated with higher 2-MIB levels, while phosphorus and nitrogen exhibited threshold-dependent relationships. SHAP analysis further demonstrated nonlinear and interactive effects among the factors. These findings indicate that AutoML provides both accurate prediction of odor risks and valuable insights into their statistical associations with water quality factors, offering technical support for early warning systems and management of drinking water reservoirs.