Hybrid Machine Learning Meta-Model for the Condition Assessment of Urban Underground Pipes

Abstract

Urban water infrastructure faces increasing deterioration, necessitating accurate, cost-effective condition assessment. Traditional inspection techniques are intrusive and inefficient, creating a demand for scalable machine learning (ML) solutions. This study develops a hybrid ML meta-model to predict underground pipe conditions using a comprehensive dataset of 11,544 records. The objective is to enhance multi-class classification performance while preserving interpretability. A stacked hybrid architecture was employed, integrating CatBoost, LightGBM, XGBoost, AdaBoost, TabNet, ANN, Logistic and Linear Regression, and Random Forest models. Following data preprocessing, feature engineering, and correlation analysis, the meta-model achieved 96.6% accuracy, outperforming individual models. Age emerged as the most influential feature, followed by material type and pipe length. ROC-AUC scores exceeded 0.95 across classes, confirming high discriminative capability. This work demonstrates the superiority of hybrid architectures for infrastructure diagnostics. Future research should incorporate real-time IoT sensor data and advanced models such as Graph Neural Networks or Transformers for dynamic, network-level condition forecasting.