Application of a predictive machine-learning model to forecast sewer’s pipes condition. A case study in Lausanne, Switzerland

Abstract

This study explores the application of a machine learning model, specifically a Random Forest classifier, to predict the condition of uninspected pipes using available structural, operational, and environmental data. Originally developed for Berlin, Germany, the model has been adapted and applied to the sewer network of Lausanne, Switzerland. Model performance was evaluated using custom metrics, with results compared to previous applications in Berlin. Despite challenges related to class imbalance, the model demonstrated promising accuracy, supporting its potential as a decision making tool for inspection prioritization.