Clustering micropollutants and estimating rate constants of sorption and biodegradation using machine learning approaches

Effluent from wastewater treatment plants is an important source of micropollutants (MPs), yet monitoring is inefficient due to their diversity. This study uses self-organizing maps (SOM) for clustering MPs and derives marker constituents to estimate behavior within clusters. Using physicochemical properties, functional groups, and initial biotransformation rules for 29 of 42 MPs, the approach ultimately estimates degradation rate constants for 13 MPs. When considering physicochemical properties and functional groups, SOM achieved labeling accuracy of 0.75 for both aerobic and anoxic conditions. Eleven MPs were identified as markers per redox condition. A random forest classifier (RFC) using these markers enabled estimation of sorption and biotransformation rate constants irrespective of dominant removal pathways, with an accuracy of 0.77 under both aerobic and anoxic conditions—exceeding prior reports. The procedure can streamline MP monitoring in WWTP effluents.