A Random Forest Algorithm Applied to Condition-based Wastewater Deterioration Modeling and Forecasting

Abstract

Condition prediction and deterioration forecasts are crucial for long-term planning of water and wastewater systems. Although sufficiently beneficial to inspire full coverage strategies, particularly in sewers, inspections are costly and it may be valuable to prioritize them based on correlations between previous results and potentially explanatory factors (e.g., age, material). The paper reports on a detailed application of a Random Forest classification algorithm in the context of a utility's thorough ongoing CCTV inspection program and AM planning, where it has been used both for estimating the importance of potential co-variates in predicting sewer condition and in directing investment in inspections.