A Comparative Study of KNN and Classification Algorithms for Wastewater Pipe Condition Rating

Abstract

Risk-based condition assessment is central to prioritizing wastewater infrastructure by linking pipe failure likelihood with associated consequences. This study presents an automated framework for assigning overall pipe condition ratings by integrating internal pipe characteristics, external environmental conditions, and hydraulic and operational indicators. Conventional manual interpretation of CCTV inspection data is labor-intensive and subject to evaluator variability, limiting scalability and consistency. To address these challenges, this work applies a K-nearest neighbors (KNN) classifier to infer comprehensive pipe condition ratings from historical inspection and attribute data. The proposed KNN model is used to identify deteriorated pipe segments requiring near-term rehabilitation or replacement. Its performance is systematically compared with decision tree (DT) and random forest (RF) classifiers using accuracy, precision, recall, and class-wise behavior across five condition categories. The framework is validated using wastewater collection system data from Shreveport, Louisiana, and is aligned with established industry condition assessment practices. Results indicate that machine-learning-based classifiers can effectively function as decision-support tools within a multi-factor pipe condition rating framework, supporting improved asset prioritization and infrastructure management.