Regularized Regression Analysis For The Prediction Of Virus Inactivation Efficiency By Chloramine Disinfection

Wastewater reclamation and reuse have been well-practiced in water-stressed areas, but insufficiently treated wastewater includes harmful contaminants. Sanitation safety planning employs the hazard analysis and critical control point to manage health risks due to waterborne pathogens including enteric viruses by determining the critical limit (CL) at critical control points (CCPs). At a wastewater treatment plant (WWTP), some disinfection conditions, such as initial disinfectant concentration, are available as parameters at CCPs when the log reduction value (LRV) of viruses is proportional to them. Since water quality affects disinfectant decay and varies among WWTPs, we have constructed models to predict virus LRVs in chloramine disinfection, in which operational and water quality parameters were used as model variables. Inactivation datasets of five viruses were collected using a systematic review method, and for model selection, we applied three regularized regression analyses (ridge, lasso and elastic net) to avoid multicollinearity. We found that lasso or elastic net regressions gave lower values of mean squared errors (MSEs) (smaller than 1, except for poliovirus), which indicated higher prediction performance. We then constructed models based on the hierarchical Bayesian approach, in which variables selected by lasso or elastic net regressions were applied, to take experimental errors among reports and strain-specific sensitivity to chloramine into account. The proposed modeling approach is useful for WWTP operators to determine the CL to maintain acceptable virus concentration in effluent.