Trihalomethane prediction model for water supply system based on machine learning and Log-linear regression.

Area of Science:

• Environmental Science
• Water Quality Monitoring
• Analytical Chemistry

Background:

• Laboratory analysis of trihalomethanes (THMs) is time-intensive.
• Developing predictive models for THMs using water quality parameters is essential for efficient monitoring.
• THMs are a group of disinfection byproducts that require careful management in water treatment.

Purpose of the Study:

• To explore and compare the efficacy of Random Forest Regression (RFR), Support Vector Regression (SVR), and Log-linear regression models for predicting THM concentrations.
• To identify the best modeling approach for estimating total THMs (T-THMs), bromodichloromethane (BDCM), and dibromochloromethane (DBCM) using nine water quality parameters.
• To establish a reliable method for monitoring THMs in water supply systems.

Main Methods:

• Development and testing of RFR, SVR, and Log-linear regression models.
• Utilized a dataset of 175 water samples from a water treatment plant.
• Input variables included nine easily obtainable water quality parameters.

Main Results:

• The RFR model demonstrated superior performance in predicting T-THMs and DBCM concentrations.
• The SVR model showed slightly better predictive accuracy for BDCM compared to the RFR model.
• RFR achieved high prediction accuracy for T-THMs (82-88%) and BDCM (85-98%) with strong correlation coefficients.

Conclusions:

• The RFR model is the most effective and overall superior method for predicting THM concentrations compared to SVR and Log-linear models.
• The developed RFR model provides a viable and efficient tool for the routine monitoring of THMs in water supply systems.
• Predictive modeling using water quality parameters can significantly reduce the reliance on time-consuming laboratory analyses for THMs.