Assessing Environmental Complexities of Decentralized Wastewater Infrastructure Using Machine Learning: A Case Study

Onsite wastewater treatment systems (OWTSs), such as septic systems, are widely used in the United States, with 16.4% of households relying on them. OWTSs process approximately 4 billion gallons of wastewater per day, yet only about half is safely treated. Identifying factors contributing to impaired functionality is crucial for developing effective management and monitoring strategies and protecting environmental and human health. This study uses a machine learning approach and a unique data set from Athens-Clarke County, Georgia, to predict OWTS failures based on environmental and system-specific variables. The three main predictors of impaired OWTS function were the number of bedrooms (25.4%), height above stream (18.6%), and system age (16.2%), with both older and younger systems prone to failure. Our findings suggest there is a need to reevaluate construction guidelines for effective tank and drainfield sizing, placement, and construction, and our findings indicate that additional training for permitters, installers, and homeowners may be beneficial. Our work demonstrates the power of using machine learning to assess OWTS function, which can better enable local governments and environment managers to identify areas in need of infrastructure and educational investment with limited data and highlights the data types that local jurisdictions should prioritize for collection.