Related Experiment Video
Updated: Jul 10, 2026
08:27
Author Spotlight: Efficient Image Recognition Using Directional Gradient Histogram Technique and Support Vector Machines
Published on: January 5, 2024
1.7K
Novel Sewer Defect Prediction Leveraging Advanced Machine Learning (ML) Models
Vannary Seng1, Barbara J Lence1, Sudhir Kshirsagar2
1Department of Civil Engineering, University of British Columbia, Vancouver, British Columbia, Canada.
Summary
Artificial intelligence (AI) and machine learning (ML) models predict sewer pipe defects, outperforming traditional methods. LightGBM models using pipe location and age data showed the best accuracy in identifying infiltration and structural issues.
Area of Science:
- Environmental Engineering
- Computer Science
- Data Science
Background:
- Traditional sewer network assessment relies on criteria-based evaluations, often failing to pinpoint specific defect locations.
- Accurate prediction of infiltration and structural defects is crucial for effective sewer maintenance and management.
- Existing methods struggle with the complexity and volume of data generated from sewer inspections.
Purpose of the Study:
- To develop and compare artificial intelligence (AI)/machine learning (ML) models for predicting sewer pipe defects.
- To assess the effectiveness of decision tree-based ML models in identifying infiltration and structural defects.
- To create utility-specific models using closed-circuit television (CCTV) inspection data and pipe information.
Main Methods:
- Comparative analysis of four decision tree-based ML models, including LightGBM.
- Development of utility-specific models using CCTV data, pipe information, and inspection reports.
- Addressing class imbalance with three methods and optimizing hyperparameters via repeated stratified k-fold cross-validation grid search.
- Performance evaluation using area under the receiver operating characteristics (AUC-ROC) and precision recall (AUC-PR) curves.
- Application of SHapley Additive exPlanations (SHAP) to identify key predictive features.
Main Results:
- LightGBM-based models demonstrated superior performance in predicting both infiltration and structural defects for both utilities.
- The best performing model achieved an AUC-ROC of 0.79 and an AUC-PR of 0.62.
- SHAP analysis identified 'pipe location' and 'pipe age' as the most significant features for defect prediction.
Conclusions:
- AI/ML offers a powerful, data-driven approach for precise sewer defect prediction, surpassing traditional assessment methods.
- The LightGBM model, particularly with cost-sensitive learning, is highly effective for sewer network defect identification.
- Pipe location and age are critical factors influencing sewer pipe integrity and defect occurrence.
