Answers are grounded in this paper and cite their sources.
Water utilities employ reactive (post-failure) or proactive (preemptive) approaches for pipeline rehabilitation. Proactive planning relies on accurate pipeline deterioration data, which is costly to collect directly. Statistical and machine learning models that predict failures with limited data are thus valuable for prioritizing rehabilitation. This study implemented two cost-sensitive binary classification models—logistic regression and support vector classification—to predict pipe failures and explored enhancing their performance by integrating K-means clustering. Two evaluation perspectives (temporal and non-temporal) were used to assess the models. Results showed that combining classification models with K-means clustering, using suitable hyperparameters, improved performance in both cases. Clustering prevented more failures by replacing fewer pipes, with a ~3% improvement in the precision-recall curve area and a 0.6-1.3% increase in F1-score. Larger training datasets also empirically enhanced model performance, aligning with expectations.
No takes yet. Share an insight, caveat, or question.
Salehinasab et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: