Machine learning-based factor analysis and risk assessment of sewage infrastructure deterioration

Abstract

Sewerage systems are critical infrastructure that support public health and water quality protection, but they face challenges from pipeline aging and low inspection rates due to limited human and financial resources. Accordingly, (1) promoting understanding of deterioration mechanisms and (2) improving the efficiency of maintenance and management are urgent tasks. In light of this, the present study aims to contribute to (1) and (2) through a three-step approach. Specifically, the approach consists of: Step 1 — identifying deterioration factors by combining a Survival Tree method (which integrates machine learning and survival analysis) with SHAP values for result interpretation; Step 2 — predicting the survival times of all pipelines under an assumed stage-wise (progressive) deterioration process; and Step 3 — prioritizing interventions using hotspot analysis based on the Getis-Ord Gi* statistic. A case study using real data from Osaka City was conducted to demonstrate this approach.