Predicting the structural condition of sewer pipes: a comparative analysis of Random Forest and logistic regression models

Ensuring the long-term sustainability of wastewater infrastructure requires proactive management. This involves identifying deteriorated pipes in the sewer network that may pose risks, enabling effective planning of rehabilitation interventions. However, conducting Closed-Circuit Television (CCTV) inspections is costly and resource-intensive, limiting their use to a small portion of the network. The present study aims to model and predict the structural condition of the pipes in the sewer network of the city of Bejaia. To achieve this, we applied binary logistic regression and Random Forest models. For the logistic regression model, we incorporated interaction terms to capture the complex relationships between the predictors and the outcome. For the Random Forest model, we tested different cutoff thresholds to optimize classification performance and determine the optimal decision boundary for predicting pipe deterioration. The results have shown that the Random Forest model demonstrates superior overall performance (70%) compared to the logistic regression model (59%).