Development of Removal Efficiency-Driven Multiple Regression Models for 2-Mib and Geosmin Prediction

2-Methylisoborneol (2-MIB) and geosmin, two primary taste and odor compounds in drinking water, cause significant quality concerns due to their algal origin and extremely low odor thresholds. This study aims to develop multiple linear regression (MLR) models to estimate 2-MIB and geosmin concentrations using total organic carbon (TOC) and chlorophyll-a as explanatory variables. Experimental data were obtained from water samples collected from Altınapa Dam and treated through various processes, including activated carbon adsorption, sulfonic acid-modified activated carbon adsorption, integrated activated carbon and peroxone process, and integrated sulfonic acid-modified activated carbon and peroxone process. Both classical linear regression and log-log models based on natural logarithmic transformations were constructed and comparatively evaluated. The results showed that sulfonic acid-modified activated carbon exhibited higher removal efficiencies for TOC, chlorophyll-a, 2-MIB, and geosmin compared to unmodified activated carbon. Maximum removal efficiencies of 91% for 2-MIB and 83% for geosmin were achieved using 8 mg/L of sulfonic acid-modified activated carbon combined with a 0.3 peroxone ratio. The log-log regression models yielded lower root mean square error (RMSE) and mean absolute percentage error (MAPE) values compared to classical linear models, indicating improved predictive capability. TOC was identified as a consistently significant predictor, while chlorophyll-a had a stronger influence on 2-MIB prediction and a more limited effect on geosmin. The findings suggest that statistical regression modeling—particularly log-log approaches—can serve as effective decision-support tools for optimizing treatment strategies to control taste and odor events in drinking water systems.