Comparison of Adaptive Neuro-Fuzzy Inference System (ANFIS) and Gaussian Process for Machine Learning (GPML) Algorithms for the Prediction of Norovirus Concentration in Drinking Water Supply

Abstract

Monitoring of Norovirus in drinking water supply is a complicated, rather expensive, process. Norovirus represent a leading cause of acute gastroenteritis in most developed countries. Modeling of general microbial occurrence in drinking water is a very active field of study and provides reliable information for predicting microbial risks in drinking water. In this work, adaptive neuro-fuzzy inference system (ANFIS) and Gaussian Process for Machine Learning (GPML) are proposed as predicting models for the total number of Norovirus in raw surface water in terms of water quality parameters such as water pH, turbidity, conductivity, temperature and rain. The predictive models were based on data from Nødre Romrike Vannverk water treatment plant in Oslo, Norway. Based on the model performance indices used in this study, the GPML model showed comparable accuracy to the ANFIS model. However, the ANFIS model generally demonstrated more superior prediction ability of the number of Norovirus in drinking water, with lower MSE and MAE values relative to the GPML model. In addition, the ability of the ANFIS model to explain potential effects of interactions among the water quality variables on the number of Norovirus in the raw water makes the technique more efficient for use in water quality modeling.