Sewer Pipe Condition Assessment Using Machine Learning: An ANN Methodology Approach

This paper introduces an innovative ensemble of eight distinct artificial neural networks (ANNs) to enhance sewer pipe condition assessment, utilizing physical and environmental data to predict structural integrity. Trained individual ANN models were averaged in the ensemble, which demonstrated a significant performance increase. The ensemble model achieved an accuracy of 0.739, exceeding the individual ANN models' average accuracy (0.698) by approximately 5.9%. The ensemble also attained a precision of 0.649, a recall of 0.739, and an F1-score of 0.689. This approach offers a reliable tool for infrastructure management, enabling accurate condition predictions and optimized maintenance strategies, contributing to smart city development and machine learning applications in sewer system evaluation, with future work planned for geographically diverse systems and temporal degradation analysis.