Scholarlink

<p>The aging sewer infrastructure in the U.S., covering 2.1&#xa0;million kilometers, encounters increasing structural issues, resulting in around 75,000 yearly sanitary sewer overflows that present serious economic, environmental, and public health hazards. This study evaluates machine learning models for reinforced concrete pipe condition prediction using artificial neural networks (ANN) and multiple linear regression (MLR) with Wisconsin sewer network data. The methodology incorporates group-based cross-validation to prevent data leakage, comprehensive feature importance analysis, and statistical significance testing across multiple pipe segments. Performance evaluation utilized classification metrics including accuracy, precision, recall, F1-score, and Root Mean Square Error (RMSE). Basic ANN achieved the highest precision of 83.4%, while Stacking Ensemble demonstrated superior overall performance with accuracy of 72.8%, precision of 82.8%, recall of 72.8%, F1-score of 75.9%, and RMSE of 0.5334. Feature importance analysis revealed that the top four predictors—age, length, pipe diameter, and depth—collectively contributed 81.6% influence on condition prediction, while other parameters provided an additional 18.4% combined influence. Statistical testing confirmed performance differences between ensemble methods and traditional approaches, validating advanced machine learning effectiveness for infrastructure condition assessment applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Machine learning models for reinforced concrete pipes condition prediction: the state-of-the-art using artificial neural networks and multiple linear regression in a Wisconsin case study

  • 期刊论文

  • 发表日期：

    2026年1月16日

查看全文
  • Mohsen Mohammadagha,
  • Mohammad Najafi,
  • Vinayak Kaushal,
  • Ahmad Jibreen

摘要

The aging sewer infrastructure in the U.S., covering 2.1 million kilometers, encounters increasing structural issues, resulting in around 75,000 yearly sanitary sewer overflows that present serious economic, environmental, and public health hazards. This study evaluates machine learning models for reinforced concrete pipe condition prediction using artificial neural networks (ANN) and multiple linear regression (MLR) with Wisconsin sewer network data. The methodology incorporates group-based cross-validation to prevent data leakage, comprehensive feature importance analysis, and statistical significance testing across multiple pipe segments. Performance evaluation utilized classification metrics including accuracy, precision, recall, F1-score, and Root Mean Square Error (RMSE). Basic ANN achieved the highest precision of 83.4%, while Stacking Ensemble demonstrated superior overall performance with accuracy of 72.8%, precision of 82.8%, recall of 72.8%, F1-score of 75.9%, and RMSE of 0.5334. Feature importance analysis revealed that the top four predictors—age, length, pipe diameter, and depth—collectively contributed 81.6% influence on condition prediction, while other parameters provided an additional 18.4% combined influence. Statistical testing confirmed performance differences between ensemble methods and traditional approaches, validating advanced machine learning effectiveness for infrastructure condition assessment applications.